ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

Contents lists available at ScienceDirect

ISPRS Journal of Photogrammetry and Remote Sensing

journal homepage: www.elsevier.com/locate/isprsjprs

Segmentation for Object-Based Image Analysis (OBIA): A review of T

algorithms and challenges from remote sensing perspective
Mohammad D. Hossain, Dongmei Chen
⁎

Laboratory of Geographic Information and Spatial Analysis, Department of Geography and Planning, Queen’s University, Kingston, ON K7L 3N6, Canada

ARTICLE INFO ABSTRACT

Keywords: Image segmentation is a critical and important step in (GEographic) Object-Based Image Analysis (GEOBIA or
OBIA OBIA). The final feature extraction and classification in OBIA is highly dependent on the quality of image
Remote sensing segmentation. Segmentation has been used in remote sensing image processing since the advent of the Landsat-1
High spatial resolution satellite. However, after the launch of the high-resolution IKONOS satellite in 1999, the paradigm of image
Image segmentation
analysis moved from pixel-based to object-based. As a result, the purpose of segmentation has been changed
Geographic object
from helping pixel labeling to object identification. Although several articles have reviewed segmentation al-
gorithms, it is unclear if some segmentation algorithms are generally more suited for (GE)OBIA than others. This
article has conducted an extensive state-of-the-art survey on OBIA techniques, discussed different segmentation
techniques and their applicability to OBIA. Conceptual details of those techniques are explained along with the
strengths and weaknesses. The available tools and software packages for segmentation are also summarized. The
key challenge in image segmentation is to select optimal parameters and algorithms that can general image
objects matching with the meaningful geographic objects. Recent research indicates an apparent movement
towards the improvement of segmentation algorithms, aiming at more accurate, automated, and computation-
ally efficient techniques.

1. Introduction 2008; Blaschke, 2010). This method intends to bypass the problem of
artificial square cells as used in per-pixel method (Fisher, 1997; Burnett
Remote sensing technology has been widely used to extract land and Blaschke, 2003; Blaschke, 2010) by grouping a number of pixels
cover/use information efficiently as it has the ability to obtain data for into shapes with a meaningful representation of the objects. The aim of
a large area repeatedly (Pu and Landry, 2012). Images captured by OBIA is to address more complex classes that are defined by spatial and
earlier remote sensing sensors such as AVHRR, MSS, TM usually had hierarchical relationships within and during the classification process
pixels bigger than ground features, requiring sub-pixels or per-pixel (Lang, 2008). OBIA is usually composed of two main phases: (1) image
analysis for features mapping (Blaschke, 2010). However, after the segmentation, and (2) feature extraction and classification. The most
launch of the IKONOS (IK) satellite in 1999, the spatial resolution of basic and critical step is image segmentation (Blaschke et al., 2008;
images increased significantly (further improved in QuickBird (QB), Cheng et al., 2001; Zhang, 1997) and the accuracy of following object-
WorldView-1 (WV-1), WorldView-2 (WV-2), WorldView-3 (WV-3), based feature extraction and classification mainly depends on the
WorldView4 (WV-4), other recent sensors and UAVs). Pixel-based quality of image segmentation (Mountrakis et al., 2011; Su and Zhang,
methods used for moderate and low-resolution imagery fail to utilize 2017). Image segmentation is defined as a method of dividing an image
the spatial variation of different land covers in the high-resolution into homogeneous regions (Pal and Pal, 1993). These regions represent
images (Campbell and Wynne, 2011) as these methods do not consider land covers such as buildings, trees, water bodies, and grasslands which
neighboring pixels which are the part of the same land cover. Conse- are known as image object in GEOBIA (Costa et al., 2018; Heumann,
quently, (GEographic) Object-Based Image Analysis (GEOBIA or OBIA) 2011).
has emerged as an effective way of analyzing high spatial resolution Image segmentation has been utilized differently in different fields
images (Blaschke, 2010). (Kerfoot and Bresler, 1999; Pham et al., 2000) such as computer vision,
OBIA is an alternative to a pixel-based method with basic analysis medical imaging, and range imaging. Many methods applied to remote
unit as image objects instead of individual pixels (Castilla and Hay, sensing imageries are imported from other fields as the underlying

⁎
Corresponding author.
E-mail address: [email protected] (D. Chen).

https://doi.org/10.1016/j.isprsjprs.2019.02.009
Received 7 November 2018; Received in revised form 8 February 2019; Accepted 17 February 2019
0924-2716/ © 2019 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V. All rights reserved.
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

principle is the same (Dey et al., 2010). Many segmentation algorithms result, pixel-based algorithms are failing to provide better accuracy in
have been developed and used in different fields. In remote sensing, high-resolution image analysis (Blaschke et al., 2004; Pu et al., 2011;
hundreds of papers involving image segmentation have been published Tehrany et al., 2014). In Geographic Information Science (GIScience),
(Zhang, 2006) although many algorithms are not applicable to the the single land cover is represented as an object and further analysis is
object-based model (Davis and Wang, 2003). It is essential to identify conducted based on objects instead of pixels. Object-Based Image
algorithms available for object-based segmentation along with their Analysis (OBIA) has been defined as “a sub-discipline of GIScience
pros and cons to ascertain the efficacy of individual algorithm. devoted to partitioning remote sensing (RS) imagery into meaningful
Several previous papers have reviewed and discussed different image-objects and assessing their characteristics through spatial, spec-
segmentation techniques such as Haralick and Shapiro, 1985; Pal and tral and temporal scale” (Hay and Castilla, 2006). The primary purpose
Pal, 1993; Schiewe, 2002; Carleer et al., 2005; Shankar, 2007; Dey of OBIA is to provide a method for analyzing high-spatial resolution
et al., 2010, 2013; Blaschke, 2010; Chen et al., 2018a. The oldest two imagery by using spectral, spatial, textural and topological character-
reviews were mostly concerned with pixel-based segmentation techni- istics (Lang, 2008). OBIA incorporate both geographic information (GI)
ques such as thresholding, Markov Random Field (MRF), neural net- and remote sensing. Image analysis is also done in other disciplines,
work, and fuzzy clustering for segmenting moderate resolution images. such as computer vision, material science or biomedical imaging.
The next four reviews have focused on the applicability of previously Blaschke et al. (2004) have introduced the term ‘GEographic Object-
developed segmentation algorithms on high-resolution satellite images Based Image Analysis (GEOBIA)’ to indicate image analysis performed
and the underlying concepts of segmentation methods. The review by remote sensing scientists, GIS specialists, and environmental dis-
paper published in 2010 (Blaschke, 2010) went through the develop- ciplines. Surveys such as that conducted by Blaschke (2010) have
ment of GEOBIA, application of multiresolution segmentation (MRS), identified 145 peer-reviewed journal paper relevant to GEOBIA. How-
algorithms developed for OBIA and their future trends. The one pub- ever, the literature search carried out by Blaschke et al. (2014) reported
lished in 2013 (Dey et al., 2013) tried to identify rules for selecting over 600 relevant journal articles on the same issue which indicates
segmentation techniques from current algorithms for urban land cover that numbers have quadrupled over four years. This article also un-
mapping and provide some insight on possible methods that can be dertakes a brief literature survey using websites such as Google Scholar,
used for the same purpose. However, image segmentation is domain Web of Knowledge and Scopus (Elsevier) with the keywords of OBIA,
specific (Zouagui et al., 2004) and still a subject of ongoing research GEOBIA, segmentation, remote sensing, object-based, object-oriented,
despite currently available numerous techniques. The latest one (Chen per-parcel, and other various spelling alternatives. As demonstrated in
et al., 2018a) reviewed the emerging trends, and future opportunities of Fig. 1, the number of articles increased significantly at the same pace.
GEOBIA and image segmentation was only a small part of it. None of There are two substantial reasons for this: availability of high spatial
the previous review papers provided a particular focus and complete resolution remote sensing images and software (both commercial and
picture on object-based segmentation algorithms used in the remote open source) for implementing GEOBIA. In GEOBIA, it is assumed that
sensing community. image objects produced by segmentation can be explicitly linked to the
This paper reviews the current object-based segmentation algo- geographic objects of interest (Shackelford and Davis, 2003; Zhou et al.,
rithms and has tried to identify their pros and cons. Peer-reviewed 2007). Thus, segmentation is the key to the GEOBIA (Lizarazo and
journals published between 1999 (when IK satellite launched) and 2018 Elsner, 2011).
have been searched using websites such as Google Scholar, Web of
Knowledge and Scopus (Elsevier) with the keywords of segmentation, 3. Segmentation
remote sensing and object-based. This review found 290 articles that
matched with the keywords and the majority of them are published in As mentioned in the Introduction section, the objective of segmen-
the following top remote sensing journals; Photogrammetric Engineering tation is to partition an image into a set of disjointed regions that are
& Remote sensing, ISPRS Journal of Photogrammetry and Remote Sensing, different according to specific properties such as texture, color, shape,
Remote Sensing of Environment, International Journal of Remote Sensing, size and gray level (Lucchese and Mitray, 2001). Mathematically, seg-
International Journal of Applied Earth Observation and Geoinformation, mentation can be defined as follows (Cheng et al., 2001): P() is the
IEEE Journal of Selected Topics on Applied Earth Observations and Remote homogeneity criteria, R is the entire image and {Ri} will be a segment
Sensing, IEEE Transactions on GeoScience and Remote Sensing, Remote of R if: (1) Ri ⊆ R (2) R = ni = 1 Ri (3) Ri Rj = (4) P (Ri Rj ) = False
Sensing, Remote Sensing Letters and Pattern Recognition. Half of the arti- when i ≠ j and Ri and Rj are neighbors. Earlier literature have cate-
cles are not included in previous review papers. Apart from those ar- gorized segmentation as (a) Pixel-based (Mardia and Hainsworth, 1988)
ticles, some earlier articles and articles from other journals have also (b) Edge-based (Perona and Malik, 1990) (c) Region-based (Beveridge
been cited for readers who are interested in understanding the funda- et al., 1989) and (d) Hybrid method (Haris et al., 1998) based on object
mental and mathematical formulation of algorithms. This review is identification method. Pixel-based methods consist of image thresh-
different from previous ones with the particular attention on object- olding and segmentation in the feature space (Schiewe, 2002). In this
based segmentation. It is also not limited to urban land cover mapping case, each spatially continuous unit needs to be assigned a unique label.
only, and it tries to indicate the strengths and weaknesses of each However, this method is not suitable for OBIA (Wang et al., 2015c) thus
method. The rest of the paper is broadly organized as follows. Section 2 it does not warrent further discussion. Apart from the earlier classifi-
presents a brief discussion on GEOBIA and OBIA. Section 3 presents cation, segmenation algorithms are also classfied based on hierarchy
different segmentation algorithms used in object-based image analysis (Guindon, 1997), object extraction method (Maxwell and Zhang, 2006),
including edge- and region-based, hybrid methods, and semantic object representation method (Rosenfield and Davis, 1979), the
techniques. The section that follows describes the challenges in seg- homogeneity criteria (Baatz and Schäpe, 2000). Besides, Zhang (1997)
mentation methods. Finally, it provides a summary of current issues. classified segmentation algorithms as the boundary- and region-based
approach based on discontinuity and similarity of object areas. This
2. OBIA and GEOBIA review is following earlier classfication system based on how segments
are generated and the subsequent sections provide details of those
Methods for low-resolution image processing are based on the methods.
classification of individual pixels (Blaschke et al., 2014). In low-re-
solution images, individual pixel contains one or even multiple land 3.1. Edge-based segmentation
cover classes. By contrast, the intra-class spectral variability is sig-
nificant in the high-resolution images (Blaschke et al., 2004). As a Edge-based techniques (Haralick, 1981; Ikonomopoulos, 1982;

116
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

Fig. 1. The amount of GEOBIA literature and some associated triggers (satellites and software).

Kundu and Pal, 1986; Lu and Jain, 1989; Zhou et al., 1989) first identify candidate to link the edge pixels. Although many algorithms have been
edges and then close them by using contouring algorithms (Cao et al., proposed to identify edges and connect them to generate objects, still
2016). It assumes that between edges, the pixel properties change identifying perfect edges to create image objects is a challenge.
abruptly (Shih and Cheng, 2004). From this perspective, edges are re- Another edge-based algorithm based on mathematical morphology
garded as boundaries between objects and located where changes occur is the Watershed Transformation (WT) (Vincent and Soille, 1991).
(Martin et al., 2004). There are many algorithms available for the Mathematical morphology is efficient than traditional edge detection
identification of the object edges that capture the geometrical and algorithms (Kaur and Garg, 2011). Thus, WT becomes most popular
physical characteristics of image objects. A variety of edge detectors has edge-based segmentation methods in remote sensing community. Wa-
been used for different purposes. Jain et al. (1995) divided edge de- tershed (Meyer and Beucher, 1990; Svoboda et al., 2007) simulates
tection into three steps: filtering, enhancement, and detection. Different real-life flooding approach (Munoz et al., 2003) and transforms the
filtering methods have been proposed (Castilla et al., 2008; Chen et al., image into a gradient and identifies objects with a topographical sur-
2006; Chen et al., 1999; Kerem and Ulusoy, 2013) to produce minimum face (Mezaris et al., 2004). As WT identifies segment boundary first and
blurring and displacement of edges. Enhancement accentuates pixels then creates the object, we consider it as an edge based method which is
where there is a significant change in local intensity values (Fosgate supported by earlier research (Carleer et al., 2005; Dey et al., 2010;
et al., 1997). Enhanced data are used for detecting real edges. Many Yang et al., 2017) as well. However, De Smet and Pries (2000) and
hard-coded operators (Andrey and Tarroux, 1998; Canny, 1987; Meinel et al. (2004) indicated that the WT has the properties of both
Deriche, 1990; Ďurikovič et al., 1995; Farid and Simoncelli, 1997; He edge detection and region growing techniques. Performance of wa-
et al., 2014; Kass et al., 1988; Leymarie and Levine, 1993; Marr and tershed segmentation largely depends on the algorithm executed to
Hildreth, 1980; Mayunga et al., 2007; Meer and Georgescu, 2001; Peng compute the gradient. Typical gradient operator produces an over-
et al., 2005; Prewitt, 1970; Robinson, 1977) are employed for edge segmented result in watershed segmentation due to noise or texture
detection. Apart from those, scholars’ also implemented soft computing pattern (Zuva et al., 2011). Although multiple algorithms (Chen et al.,
methods such as fuzzy-based approach (Trivedi and Bezdek, 1986), 2008, 2006; Sun and He, 2008; Tarabalka et al., 2008; Wang, 1997;
genetic algorithm-based approach (Natowicz et al., 1995), and neural Weickert, 2001) proposed to generate the gradient image in order to
network-based approach (Manjunath et al., 1990). Nevertheless, all increase the performance of the WT, still watershed generate an over-
operators create broken edges and miss some essential edges (Jevtic segmented image. Nevertheless, among the edge-based techniques, the
et al., 2009). The performance of each operator is evaluated based on watershed is widely used for natural image segmentation. Table 1 de-
the false edge, missing edge, edge angle, distance from the true edge monstrates some algorithms used for watershed segmentation. Only few
and distortion (Lucchese and Mitray, 2001). studies found in literature where a solely edge-based method was used
After the identification of edges, the next step is transforming the for object-based segmentation in remotely sensed images.
edges into closed boundaries. This step often involves the exclusion of
edges that are produced by noise, the connecting of gaps at places 3.2. Region-based segmentation
where no edge is detected and decisions to join those edge parts that
make up a single object. Multiple edge-linking methods (Jaing et al., The edge-based methods try to explore the object boundaries first
1994; Liu et al., 2008; Lu and Chen, 2008) have been suggested to and then determine the object itself by filling them in (Guindon, 1997).
compensate edges that are not entirely linked. Scholars also utilize The region-based methods (Davis et al., 1975; Ohta et al., 1980; Pal and
Hough transform (Ballard, 1981; Kiryati and Eldar, 1991) to search the Pal, 1987; Pong et al., 1984) take the opposite approach. Those start
ideal edges that best fit the partial edges. However, this method works from the inside of an object and then expand outward until meeting the
well for simple parametric shapes (Maintz, 2005). Researchers also object boundaries (Zhang, 2006). Theoretically, edge-based and region-
utilized neighborhood search (Ghita and Whelan, 2002) to find a based are different representations of the same object. However, the

117
 M.D. Hossain and D. Chen

Table 1
Summary of watershed-based segmentation techniques.
Sub-Class Procedure Performance and advantages Limitations Test data (sensor, spatial resolution, Previous literatures
bands) and application

Edge-constrained Edge-constrained watershed segmentation and • Could able to reduce false edges • Not suitable for strongly • Aerial, SPOT and ALOS Wang and Li (2014)
edge allocation were used to generate sub- caused by noise and formed textured images. • 0.3 m, 10 m and 2.5 m respectively Others:
object primitives boundary-closed segments. • Require additional • RGB, RGB + NIR and panchromatic Wang et al. (2015b)
• Provided high segmentation accuracy computation time. (PAN) respectively
and boundary precision. • Target: not specified
Marker An Edge Embedded Marker-based Watershed • Successfully segmented small objects • Parameters were chosen • QB Li et al. (2010a)
(EEMW) algorithm was utilized for marker and provided accurate boundary. manually. • No details for resolution, and bands Others:
extraction and pixel labeling • Integrated edge information into • Target: road, green belt, playground, Gaetano et al. (2012), Jiao et al.
segmentation. farm, lake, and house (2010), Turker and Sumer (2008), Li
et al. (2012), Gaetano et al. (2015)
Edge-embedded The result of the Canny edge detection was • Provided initial segments. • Broken and burrs roads • ALOS, GF-1 Wang and Wang (2016)
embedded into watershed segmentation region boundary achieved by this were in the results. Others:

118
• The
method was highly consistent with further processing
• 10RGBm+andNIR8 m respectively Mylonas et al. (2015)
actual boundaries.
• The
required to improve
• Target: road
region shape analysis and accuracy.
•
• Improved
spatial relationship reasoning.
Grey Level Co-occurrence In addition to the gradient, GLCM was also • Provided different accuracy for road single scale used for • QB Wang et al. (2005)
Matrix (GLCM) added as a separate band in watershed and farmland.
• Asegmentation. m multispectral (MLS)
segmentation reduced for still
• 2.44
• Over-segmentation
farmland.
• Over-segmentation
exists.
• RGB
texture of the image object.
• Target: road and agricultural land
Classification-based Watershed segmentation utilized an inverted
• Utilized
selection was the key. only on binary image Levner and Zhang (2007)
probability map as an input for flooding
• Seed generate object efficiently.
• Applicable
classification.
• Aerial
details provided for resolution and Others:
• Can an adaptive object
• No
bands Derivaux et al., (2010)
• Implemented
recognition framework. frozen oil, sand ore, mixed
to multi-channel data.
• Target:
vegetation
Hierarchical Multilevel hierarchical segmentation was
• Applicable better than single-level only to map and IK Li et al. (2011)
created using the gradient generated from the
• Performed
watershed segmentation.
• Applied
impervious surfaces.
• QB m (MLS) and 0.61 m (PAN) for Others:
multichannel morphological technique
• 2.44
QB, 4 m (MLS) and 1 m (PAN) for IK. Najman and Schmitt (1996)
• Reduced miss-segmentation errors. + NIR + PAN
• RGB
• Target: impervious surface
ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

region-based approaches may generate radically different results than known as pairwise data clustering (Blaschke et al., 2008). Instead of
edge-based approaches (Kavzoglu and Tonbul, 2017; Maintz, 2005). global criteria, it uses local criteria and can integrate spectral, shape,
Region-based methods assume that neighboring pixels within the same texture, size, spatial, prior knowledge and contextual properties of the
region have similar values (Tremeau and Borel, 1997). Region-based image objects.
segmentation methods have two basic operations: merging and splitting Apart from the MRS, researchers also implemented other region
(Fan et al., 2001). The basic approach to region-based image segmen- merging approaches such as Mean-Shift (MS) (Comaniciu and Meer,
tation is (Bins et al., 1996): (a) obtain an initial (over or under) seg- 2002), Hierarchical Stepwise Optimization (HSWO) (Beaulieu and
mentation of the image, (b) merge or split those adjacent segments that Goldberg, 1989), and Recursive Hierarchical Segmentation (RHSeg)
are similar (or dissimilar) and (c) go to the previous step until no seg- (Tilton et al., 2012). The MS is a clustering algorithm with non-para-
ments that should be merged or split remain. metric density estimation. It segments the image by grouping all pixels
that are closer in the spatial and spectral domain and then connecting
3.2.1. Region growing/merging the corresponding convergence points. Based on this aspect, this review
Region growing (Adams and Bischof, 1994) is the most popular and considered MS as a region-based approach. HSWO is also a clustering
simple algorithms for region-based segmentation. There are two main method that starts with individual data point and sequentially reduces
issues in region merging/growing segmentation, selection of seed re- number of clusters by merging. It utilizes a heap data structure to de-
gion and similarity (Lucchese and Mitray, 2001). Seeded region termine best merge. Hierarchical Segmentation (HSeg) (Tilton et al.,
growing (Deng and Manjunath, 2001) has two internal pixel order 2010) is the improvement of HSWO in the merging process. It in-
dependencies that create different resulting segments (Mehnert and corporates spectrally similar but spatially disjoint regions in the mer-
Jackway, 1997). If multiple pixels have the same difference measure to ging step. This process requires excessive computation time. To over-
their neighboring regions, then first order dependency happen (Shih come this issue, RHSeg divides the image into subsections and apply
and Cheng, 2005). The second-order dependency arises when one pixel HSeg on each subsection. Finally, it recombines the segmentation re-
has the same variation measure in several regions. Seed selection in- sults from the subsections. This aggregation method may lead to pro-
crease computational cost and execution time (Freixenet et al., 2002). cessing window artifacts. Therefore, RHSeg contains an additional step
To overcome the seeding problem, a single-seeded region growing to eliminate the artifacts.
technique was proposed by Verma et al. (2011). Mirghasemi et al.
(2013) utilized Particle Swarm Optimization (PSO) to solve the locali-
zation problem. Wang et al. (2010) implemented the K-means clus- 3.2.2. Region splitting and merging
tering algorithm to generate seed in Region-based Image Segmentation As global measures (used in region merging) caused imbalanced
Algorithm (RISA). Zhang et al. (2014a) proposed a hybrid region performance (Yang et al., 2017), researchers implemented split and
merging (HRM) method to segment high-resolution remote sensing merge (Cheevasuvit et al., 1986; Chen and Pavlidis, 1979; Chen et al.,
images. They combined global-oriented and local-oriented region 1991; Horowitz and Pavlidis, 1976; Suk and Chung, 1983) as local
merging strategies into an integrated framework. By contrast, Byun measures for segmentation process in order to get a better result (Liu
et al. (2011) presented an approach based on modified seeded region and Sclaroff, 2001). The splitting process starts from the entire image
growing and region merging utilizing a block-based seed selection then based on the criterion for inhomogeneity (grey values, texture,
method. Researchers are still in search of the better approach that can internal edges, or various other criteria) split the image into segments
serve without seed (Wu et al., 2015) or can be unbiased by neighbors (Blaschke et al., 2004). The split and merge method combines a bottom-
even though seeded (Fan et al., 2005). up approach with a top-down approach (Guindon, 1997). Bottom-up
After the selection of seeds, the region then grows by adding ad- approach generates an object by combining/merging pixels while the
jacent pixels that are similar, according to a specific homogeneity cri- top-down approach shifts from splitting the entire image into image
terion, increasing the size of the region gradually. Therefore, the objects based on heterogeneity criteria (Benz et al., 2004). In this case,
homogeneity criterion is the critical function of determining whether a the initial seed is merely the entire image (Ohlander et al., 1978). If the
pixel belongs to the growing region or not (Nock and Nielsen, 2004). seed is not homogeneous, the splitting method divides the seed into
The decision to merge is based only on the contrast between the current four squared subregions. Those subregions act as a seed in the next level
pixel and the region. Falah et al. (1994) and Xiaohan et al. (1992) and continue until all subregions become homogeneous (Martin et al.,
implemented a homogeneity criterion containing the value of the 2004). Kelkar and Gupta (2008) introduced improved quadtree method
modulus of the gradient of the pixel and the weighted sum of the for the split-and-merge segmentation. Manousakas et al. (1998) im-
contrast between the region and the pixel. Only pixels having low plemented principles of simulated annealing and boundary elimination
gradient values (below a certain threshold) were aggregated to the to improve the quality of traditional split and merge algorithms.
growing region in each iteration. In those cases, the spatial resolution of However, the fusion of two segments is upfront, the splitting of a seg-
the image played a significant role. ment requires proper sub-segments. The primary disadvantage of re-
Any object that is smaller than the spatial resolution of the image gion splitting is that the resulting image tends to mimic the data
cannot be detected in segmentation process. However, if objects are structure used to represent the image and comes out too square (Cheng
bigger than spatial resolution, then it will be fragmented onto pixels. In et al., 2001). The region merging approach is often combined with
contrast to the region growing techniques, region-merging techniques region splitting to merge the similar regions for creating a homo-
start from an initial region. Multi-Resolution Segmentation (MRS) arose geneous region as large as possible. Alshehhi and Marpu (2017) exe-
from this idea (Hay et al., 2003). The Fractal Net Evolution Approach cuted hierarchical merging and splitting image segmentation based on
(FNEA) is a multiresolution approach developed by Baatz and Schäpe color, and shape features for road extraction from urban area images.
(2000) and implemented in many research (Gao et al., 2017; Johnson, Table 2 summarized different region-based segmentation techniques.
2013; Kalantar et al., 2017; Li et al., 2016; Mallinis et al., 2008; As indicated in the table, most algorithms are facing difficulty in de-
Mathieu et al., 2007; Srivastava et al., 2015; Wang and Aldred, 2011; fining appropriate parameters. In addition, though some of them gen-
Yang et al., 2015c). The FNEA is a region merging hierarchical seg- erated promising results, they have not been evaluated by applying to
mentation and starts with 1-pixel (Blaschke et al., 2004). Each coarse different image setting. Another issue is that in many cases segmenta-
level acquires input from a finer level. If an object is recognized at the tion results were compared with segments generated by eCognition
finer level, then it repeats its representation at each coarse level. It (using FNEA) even though its results depend on user-defined para-
utilizes pairwise merging to form bigger objects. This procedure is also meters.

119
 Table 2
Summary of different region-based segmentation techniques.
Algorithms Sub-Class Procedure Performance and advantages Limitations Test data (sensor, spatial resolution, Previous literatures
bands) and application

Region growing/ Seeded The seed was identified based on • Work well in identifying • Further investigation is required to • UAV Zhou et al. (2016)
merging geometry using Stroke Width roads irrespective of shape, identify the applicability of the details provided for Others:
M.D. Hossain and D. Chen

Transformation (SWT). A Gaussian width, direction or intensity proposed method in images where
• No
resolution, and bands Epshtein et al., (2010), Zivkovic
Mixture Model (GMM) was variation. there are land covers with similar • Target: road (2004), Deriche et al. (2017)
implemented to differentiate • Lower running time geometric and radiometric Others for segmenting roads:
background and non-background. compared to other algorithms characteristics, such as long narrow Mohammadzadeh and Zoej (2010),
Finally, a Convex Active Contour implemented for the same turbid water canal. Sun and Messinger (2013)
(CAC) model was operated to merge purpose.
road seeds and identify whole road
segments
The seed was generated from the • Performed fine in segmenting • Manual selection of parameters is • QB, IK, and WV-2 Liu et al. (2015)
gradient image. Spectral- building and other fabricated required. • 2.4 m (MLS) and 0.6 m (PAN) for Others:
morphological characteristics of a structures in urban and QB; 4 m (MLS) and 1 m (PAN) for Pesaresi and Benediktsson (2001),
pixel were considered as a criterion of suburban areas. IK; 2 m (MLS) and 0.5 m (PAN) Bellens et al. (2008), Barrile and
homogeneity • Utilized both spectral and for WV-2. Bilotta (2016), Ling et al. (2012),
spatial information of images. • RGB + NIR + PAN for QB and IK; Yuan et al. (2014), Wang et al.
• Automated seed selection and nine bands for WV-2 (2015c)
merging process. • Target: structures in urban and
suburban areas.
The seed was the central pixel of each addition to merging • Manual selection of parameters is • GF-2 Su (2017)
square if image tessellated. Local
• Incriteria (MC), it provided required. 4m Others:
mutual best fitting (based on spectral importance to merging order
• RGB + NIR Tilton et al. (2012)
variance and inter-segment edge (MO) as well.
• Target: building, road,
strength) rule was used to identify Merging priority was
• agricultural land

120
appropriate neighbors for merging
• estimated based on the inter-
segment heterogeneity and
intra-segment homogeneity.
Mean Shift Horizontal and vertical semivariogram • The proposed method could • Applied only on PAN images. Ming et al. (2012)
was used to identify spatial bandwidth identify the appropriate scale. research is required to
• IK1 mandfor QBIK and 0.7 m for QB Others using semivariogram:
(window size). Acquired bandwidth Utilized global and local
• Further
evaluate the performance in
• PAN Karl and Maurer (2010)
was applied on mean-shift-based
• structure of the objects in segmenting nested structures.
• Target: cropland
multiscale segmentation multiple scales.
•
• Combined geostatistics and
pattern recognition.
Overlapping tiles were generated to • Successfully remove artifacts • Intrinsic performance evaluation was • Pleiades Michel et al. (2015)
apply stabilized mean-shift filtering in segmentation due to tiling. not compared among different m Other tile-wise:
algorithm. Each tile was processed Processing time improved algorithms.
• 2Spectral bands not mentioned Tilton (2010), Michel et al. (2012),
independently using the connected-
• significantly using the parameters were utilized
• Target: multiple objects Banerjee et al. (2012), Körting et al.
component algorithm. Unique labeling proposed method.
• Arbitrary
during the segmentation.
• (2013), Tzotsos and Argialas (2006),
conducted by shifting values from one • Improve stability issues in the Xing et al. (2014)
tile to another mean-shift segmentation Others using mean shift:
algorithm. Wang et al. (2015a), Wang et al.
• Tile-wise segmentation (2012)
overcomes the issue of
segmenting a large dataset.
Region The local best region-growing strategy • Initial over segments ensure • Though the value range and physical • QB, WV-2, and aerial Zhang et al. (2013)
Adjacency Graph was used to create initial segments. inclusion of all object meaning are different, the sum of • 0.6 m (pansharpened), 2.0 m and Others:
(RAG) RAG was created based on initial boundary. standard deviation and compactness 0.2 m respectively. Yu and Clausi (2008), Zhang et al.
segments. Edge strength was used as a • RAG indicated the were applied as the merging criterion. • RGB + NIR for QB; eight bands (2015a,b,c,d), Sarkar et al. (2000)
merging criterion. In order to produce relationship between the • The scale was predefined. for WV-2; RGB for aerial.
multi-scale segments, a local best neighboring segments. • Target: settlements, road, pond,
region merging process was applied • Edge strength ensures farmland, forest.
merging the right segments.
(continued on next page)
ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134
 Table 2 (continued)

Algorithms Sub-Class Procedure Performance and advantages Limitations Test data (sensor, spatial resolution, Previous literatures
bands) and application

Statistical sorting Initial segments were generated using • Utilized spectral, spatial, • Need multiple user inputs. • QB Li et al. (2009)
statistical region merging and scale, and shape of image • Success depends on sort function and • 2.44 m (MLS) and 0.61 m (PAN) Others:
minimum heterogeneity rule was objects. merging predicate. + NIR + PAN Huang et al. (2014), Nielsen and
M.D. Hossain and D. Chen

utilized for object merging small redundant

• RGB road, highway, grass, and Nock (2003)
• Eliminated
objects.
• Target:
buildings.
• Generated results in vector
and raster format.
Multi features The scale was selected based on local • The sequence of multi-scale J- • Tested only on a subset of the image. • QB Wang et al. (2016)
area homogeneity index J-value. Inter- images can adequately reflect • 2.4 m spatial resolution Others:
scale boundaries constraint strategy homogeneity of spectral • RGB + NIR Cánovas-García and Alonso-Sarría
was used for multi-scale segmentation. distribution of local region. • Target: road, playground, water, (2015)
Merging was done based on multi- and artificial targets
features
Structural Structural constraints such as parallel • Utilized straight-line • Applicable to man-made structures • Aerial and ALOS Wang et al. (2017a,b,c)
constraints straight-line neighborhood, boundaries featured for roads only. • 0.3 m for Aerial and 2.5 m for Others:
perpendicular straight-line and buildings. ALOS Wang et al. (2017c), Wang et al.
neighborhood, and parallel straight- • RGB for aerial and PAN for ALOS (2017a,b), Wang et al. (2015a,b,c),
line zone were incorporated in the • Target: buildings and roads Wang and Li (2014) Wang and Wang
merging process (2016)
Scale-space Anisotropic morphological leveling • The segmentation algorithm • Accuracy was assessed for • Aerial, SAR and Landsat Tzotsos et al. (2011)
filtering was used for removing noise and was applied to simplified classification results only. • Spatial resolution not specified Others:
complexity available in high- images. • Panchromatic and multispectral Kavzoglu et al. (2016)
resolution images. MSEG (Tzotsos and • Get rid of tuning shape, color • Target: building, impervious
Argialas, 2006) was employed for and texture parameters. surface, woodland, grassland,
generating primitive image objects waterbodies, etc.

121
Split and merge Spectral variance Initial segments were generated based • Provided better results in • Five parameters must be set manually • WV-2, aerial and RapidEye image Chen et al. (2015)
difference on spectral heterogeneity. Merging over-, under- and well- to implement the algorithm. • 0.6 m (pansharpen), 1 m and 5 m Other split and merge:
was done based on spectral variance segmentation rate. • Scale needs to be set small initially respectively Deng et al. (2013), Lucieer et al.
difference and edge penalty • Constrained spectral variance which cause over-segmentation. • RGB + NIR (2005), Hu et al. (2005), Miao et al.
difference can limit the • Required additional steps compared to • Target: farmland, road, buildings, (2015)
influence of the segment size. MRS. river, and reservoir Edge penalty:
• Edge penalty provided Zhang et al. (2014a,b,c)
accuracy in merging
boundary.
Hierarchical In the splitting phase, segments were • Identified complex textured • Initial segments were created using Gaetano et al. (2009)
multiple Markov created based on spatial and spectral areas efficiently. only PAN band.
• Ik1 m (PAN) and 4 m (MLS)
chain properties of objects. Using those a sequence of nested
• PAN, RGB + NIR
segments as a unit, merging was done
• Provided
segmentation maps.
• Manually selected parameters. • Target: roads, parking lots,
based on texture
• buildings, trees, grass, etc.
Hierarchical Split Band ratio was employed as a region • Removed non-continuous • Segmented regions into rough land • QB Wuest and Zhang (2009)
Merge description. Entropy was used as a regions. cover classes. • 2.44 m Others:
Refinement texture measurement. Fuzzy logic • Agglomerative merging • Manually selected parameters. • RGB Ojala and Pietikäinen (1999), Chen
based similarity measure was used for process reduced undesired • Target: forest, grass, water, soil and Chen (2002)
merging segments. and urban.
ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

3.3. Hybrid method (HM) 2010), Neural Network (NN) (Kurnaz et al., 2005; Villmann et al.,
2003), active Support Vector Machine (aSVM) (Mitra et al., 2004),
To overcome the limitation of both edge- and region-based method, weighted aggregation (Du et al., 2016), and Deep Convolution Neural
scholars integrated the results of edge- and region-based method and Network (DCNN) (Audebert et al., 2016) are supervised approaches. In
are expected to provide better segmentation results (Al-Hujazi and semantic segmentation method, each pixel is allotted a class label of its
Sood, 1991; Fan et al., 2001; Moigne et al., 1995). As discussed earlier, enfolding object.
edge-based methods are precise in detecting edges, however, facing Among the semantic algorithms, MRF accounts for a large percen-
problem in generating closed segments. By contrast, region-based tage (Geman and Graffigne, 1986; Melas and Wilson, 2002; Zhuowen
methods create closed regions, however, resulting in imprecise segment and Zhu, 2002). MRF is a probabilistic method that seizes the con-
boundaries (Wang and Li, 2014). As a result, a recent trend in image textual limits within the neighboring pixels. Feng et al. (2010) used
segmentation is to execute an HM (Gaetano et al., 2015; Li et al., 2014), split-and-merge techniques to segregate the main problem to a series of
in which the initial segments are first outlined using edge-based sub-problems. Tree-structured graph cut, hierarchical graph cut and
methods, then merged using region-based methods. Such HMs utilize net-structured graph cut was used to obtain labeling accuracy and
both boundary pixels to outline the initial segments and the interior spatial coherence. The proposed method was computationally efficient
pixels to merge similar segments (Zhang et al., 2014b). Mueller et al. and well performed in terms of robustness to noise and soft boundary
(2004) combined edge and region-based techniques to extract large preservation. In order to obtain proper segments from noisy images
man-made objects such as agricultural fields. In the first part, they with the complex and macro-texture pattern, Zheng et al. (2013) in-
extracted shape information. The edge map offers an additional cri- tegrated the MRF model with Multi-Region Resolution (MRR) seg-
terion in decisions. In the second part, they used this information to mentation. They applied the proposed method to QB, SPOT-5, and
control region growing algorithm. On the other hand, Gambotto (1993) Synthetic Aperture Radar (SAR) images. The proposed method seg-
suggested using edge information to stop the growing process. Li et al. mented images into three broad land cover types only such as farmland,
(2010c) recommended texture clustering which was executed as con- woodland, and urban area. Researches also utilized edge penalty
straints in HSWO. Merging was conducted based on the region ad- function (Yu and Clausi, 2008), discrete wavelet transform (Jung et al.,
jacency and neighbor graph. 2005), multiscale approach (Moser and Serpico, 2008) and region-
Many region-merging methods employ a single global parameter to based strategies (Moser et al., 2013) with MRF to segment high-re-
control the iterative process of merging segments as it gives the user solution images. Recently among the semantic methods, DCNN has
control over under- and over-segmentation. Even so, the same threshold been used in researches (Kemker et al., 2018; Nigam et al., 2018; Zhang
is used for all segments regardless of their homogeneity with other et al., 2016) as it has the capability to treat data as a nested model. In
segments. Instead of using a global threshold, Johnson and Xie (2011) those methods, raw images are used as an input and pixels generate
and Chen et al. (2014) have used local measures to identify segments object when it passes through multiple layers.
that are under and over segmented at the selected optimal scale para- Though semantic algorithms showed encouraging results, they are
meter and further refined them by appropriate splitting and merging. facing many challenges. For instance, in high-resolution images, it is
This local refinement strategy is efficient in improving segmentation difficult to define suitable features with semantic meaning due to high
quality because it eliminates under and over-segmentation problems texture (Zhang et al., 2014a). In addition, the scale and hierarchy
(Yang et al., 2017). However, the further splitting and merging steps available on those images make it difficult in determining semantic
face a challenge when executing local refinement in an operational rules (Burnett and Blaschke, 2003) which can differentiate objects in
context (Yang et al., 2016). In addition, those methods considered different scales. Apart from that, different image objects may have si-
heterogeneity between adjacent segments as the merging criteria. Both milar spectral value (such as water and shadow) which create ‘semantic
homogeneities within the segments and heterogeneities between the gap’ (Wang et al., 2013). Furthermore, semantic algorithms also suffer
segments should be considered equally. Wang et al. (2018c) proposed computational burden to extract structural information (Yang et al.,
an HM considering the objective of heterogeneity and relative homo- 2008), require a vast amount of training data and a significant number
geneity during the merging process. Table 3 compiled different hybrid of parameters for tuning (Chen et al., 2018a).
segmentation methods. Most of the studies started from the edge-based
method which create an over-segmented image. Then the region-based 3.5. Available software/tools
method was conducted to merge similar segments based on either
homogeneity or heterogeneity. In the merging process, there are two Though hundreds of algorithms have been proposed for segmenta-
main issues, merging criteria and merging order. Variance, area tion, only a few of them have been implemented and are available as a
weighted variance, Moran’s I, spectral angle, F measure, spectral and tool/software. Among them, eCognition is the popular and widely used
geometric properties were used as the merging criteria. To identify the segmentation software. According to Blaschke (2010), 50–55% OBIA
adjacent relationship, many studies utilized RAG, and nearest neighbor articles employed eCognition. Its success prompted other commercial
graph. As illustrated in the table, when combining both the edge- and software developers such as Hexagon Geospatial, Harris Geospatial
region-based methods, issues of the individual algorithm such as over- Solutions, ESRI, and PCI Geomatics to develop their tools. Apart from
segmentation, seed selection, generate intial region for merging, under- those, another set of tools such as EDISON, SCRM, and GeoSegment
segmentation are compensated by other(s). Though HMs provide some developed in an academic environment, and others are open source
promising results, their implementations are troublesome. tools (Table 4) such as SAGA, GRASS GIS developed by other devel-
opers.
3.4. Semantic methods
4. Discussions
Machine learning (ML) has proven successful for many applications
in recent years that affect remote sensing arena as well. ML algorithms 4.1. Pros and Cons of different algorithms
are “approximators” which learn from the training data and act ac-
cordingly. Unlike unsupervised methods (such as region growing), ML- Advantages of edge-based segmentation are that algorithms are less
based semantic segmentation algorithms such as Markov Random Field complicated compared to region-based segmentation (Felzenszwalb and
(MRF) (Farag et al., 2005; Krishnamachari and Chellappa, 1997; Poggi Huttenlocher, 2004), works fine in images with the decent contrast
et al., 2005; Tran et al., 2005; Tupin and Roux, 2005), Bayesian Net- between object and background (Kaganami and Beiji, 2009; Tsai et al.,
work (Bouman and Shapiro, 1994; D’Elia et al., 2003; Zhang and Ji, 2003), computationally efficient (Lin et al., 2003) and can correspond

122
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

Table 3
Summary of different hybrid segmentation techniques.
Algorithms Performance and advantages Limitations Test data (sensor, spatial resolution, Examples
bands) and application

Spectral Angle (SA), • Integrated multi-bands into the • Over-segmentation existed in final • QB Zhang et al. (2008a)
Watershed segmentation process. segments. • 0.61 m (PAN) and 2.5 m (MLS) Others:
Transformation (WT), • Combined both edge- and region- • Manual selection of threshold for • RGB + NIR and PAN Kruse et al. (1993)
and RAG based segmentation. region merging. • Target: Agricultural land
WT, threshold-based region • Provide higher accuracy than • Implemented on the moderate • Landsat-5 Thematic Mapper Yang et al. (2016)
merging traditional fixed-threshold region resolution images only. • 30Seven
m Others:
merging. • Target:spectral bands Yang et al. (2017), Liu
• Utilized self-adaptive spectral angle
for local-oriented region merging that
• cultivated farmland (2018)

overcome limitation of fixed

parameters in region merging.
WT, heterogeneity-change- • Could segment adequately even small • Supervision required during the • QB Chen et al. (2014)
based merging objects. merging process. • 0.61 m Others:
• Few parameters • RGB Chen et al. (2012a,b)
• Utilized spatial features to get optimal
scale.
• Target: buildings, pavements,
trees, grass, shadows, and water
Gravitational field-based • Provided more homogeneous regions • Overall implementation • WV-2, QB and aerial Zhang et al. (2017a)
segmentation, and precise edge localization. consumed more time than • 0.5 m (pansharpen) for both WV-2 Others:
hierarchical region • Provided better initial segments for eCognition’s multiresolution. and QB; 0.3 m for aerial. Rashedi and
merging region merging. • RGB + PAN Nezamabadi-pour (2013)
• Reduced computation time.
Edison operator, FNEA, ant • Eliminated disturbances. • Depends on global optimization • QB Yin et al. (2015)
colony optimization • Applicable to any high-resolution technique and intelligence of ants. • 0.61 m Others:
images. • RGB + NIR
Target: road
Zarrinpanjeh et al.
(2013), Miao et al.
(2013)
R-tree, RAG • Suitable for images having mixed scale • Generated over-segmented • IK2.4 m Hu et al. (2016)
objects. objects. • RGB Others:
• Eliminated the limitation of • Target: farmland, residential, Chen et al. (2009)
traditional MRS. • ponds, coastal, etc.
• Adopted
parameter.
global edge statistical

Canny edge detector, • Could successfully adjust the • Buffer and grid used for boundary • QB and ASTER Judah et al. (2014)
boundary adjustment, boundary of a segmented map. adjustment were chosen by trial • 2.4 m and 15 m Others:
MRS • Overall homogeneity within the and error. • RGB + NIR for QB, RGB for Zhang et al. (2008b)
segments increased. • Boundary pixels were processed ASTER
• Match closely with the object
boundary.
several times. • Target: multiple objects
• Integrated edge- and region-based
active contouring models.
Morphological information, • Combined spectral and morphological • Further study is required to check • QB Liu et al. (2015)
region merging characteristics together. the robustness of the proposed • 2.4 m (MLS) and 0.6 m (PAN) Others:
• Seeds were generated automatically. algorithm. • RGB + NIR Akçay and Aksoy (2008)
• Provided a better result than the use of
morphological characteristics alone or
• Target:
trees.
buildings, roads, and

MRS implemented in eCognition.

Gradient, WT, and region • Provide the initial step to convert an • Used only radiometric distance • Landsat ETM+, SPOT 5 and QB Castilla et al. (2008)
merging image into a shape-oriented (segment) between region centroids. • 252.8mm(resampled), 2.5 m and Others:
representation. • Can use as guiding template for respectively Hay et al. (2005), Hay
• Can generate segments with minimum photo interpretation. • RGB et al. (2003), Yue et al.
user input. • Target:
features
natural and semi-natural (2011)

Region merging, region • Multiple segmentation techniques • Require multiple parameters • Aerial Li et al. (2014)
splitting were used to segment different types tuning. • 1RGBm
of shapes. • Require expert knowledge to • Target:+ NIR
• Segmenting objects in different
hierarchical level provided better
select bands for segmenting
individual land cover.
• imperviouswater, grass, soil,
surface, tree and
results. agricultural land.

well with the object edges (Chen et al., 2018b). Problems in edge-based contextual information at larger scales (Gaetano et al., 2015) which is
segmentation are that algorithms do not function well on images with the key in object-based image analysis. However, the edge-based
smooth transitions and low contrast, sensitive to noise (Iannizzotto and method can be used to support region-based techniques (Sappa, 2006).
Vita, 2000). Due to poor performance in the detection of textured ob- The region-based methods generate spatially and/or spectrally
jects (Yu et al., 2006), edge-based segmentation has not been applied homogeneous segments based on the defined properties. In addition,
widely in high-resolution images. In addition, if it misses part of the region-based methods can produce segments at multi-scales. For in-
boundary, then disjointed edges permit merging of dissimilar regions stance, the shape of a segment at one scale level can be used as a
(Kermad and Chehdi, 2002). Furthermore, the multi-spectral image variable at another level (Wang et al., 2010). Also, those methods allow
makes the edge detection process more complicated (Li et al., 2010b) users to choose multiple criteria at the same time. Furthermore, users
due to the inconsistent location of edges in the multiple bands. Finally, have the freedom to select the seed point and merging criteria. Finally,
edge-based methods rely on local data and thus misses essential those methods are less sensitive to noise when compared with edge-

123
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

Table 4
Available software/tool for object-based segmentation.
Tool/Software Reference/Developer Website Algorithm Availability

InterSeg Happ et al. (2016) http://www.lvc.ele.puc-rio.br/wp/?cat=41 Region-based (on Available upon request
cloud)
SEGEN Gofman (2006) http://www.research.ibm.com/haifa/projects/image/segen/index.html Region-based Commercial
BerkeleyImgSeg Clinton et al. (2010) http://www.imageseg.com/ Region-based Commercial
Orfeo Toolbox Grizonnet et al. (2017) http://www.orfeo-toolbox.org/otb/ Region-based Freeware
RHSeg Tilton et al. (2012) https://opensource.gsfc.nasa.gov/projects/HSEG/ Region-based Evaluation copy
IMAGINE Spatial Modeller Hexagon Geospatial http://community.hexagongeospatial.com/t5/IMAGINE-Spatial-Modeler/ Edge-based Commercial
tkb-p/eTSpatialModeler
ENVI Feature Extraction Harris Geospatial https://www.harrisgeospatial.com/docs/routines-164.html Edge-based Commercial
Solutions
IDRISI GIS Tool Clark Labs https://clarklabs.org/terrset/idrisi-gis/ Edge-based Commercial
GRASS GIS Neteler et al. (2008) https://grass.osgeo.org/grass74/manuals/i.segment.html Region- and edge- Freeware
based
Object Analyst PCI Geomatics http://www.pcigeomatics.com/geomatica-help/concepts/focus_c/oa_intro. Region-based Commercial
html
eCognition Developer Baatz and Schäpe (2000) http://www.ecognition.com/suite/ecognition-developer Region- and edge- Commercial
based
SPRING Câmara et al. (1996) http://www.dpi.inpe.br/spring/english/index.html Region- and edge- Freeware
based
EDISON Comaniciu and Meer, http://coewww.rutgers.edu/riul/research/code/EDISON/doc/help.html Region-based Freeware
(2002)
SCRM Castilla et al. (2008) http://www.castlink.ca/scrm/scrm Region- and edge- Freeware
based
RSGISLib Bunting et al., (2014) https://www.rsgislib.org/ Region-based Freeware
SAGA Böhner et al., (2006) http://www.saga-gis.org/en/index.html Region- and edge- Freeware
based
Feature Analyst Opitz and Blundell, (2008) https://www.textronsystems.com/what-we-do/geospatial-solutions/feature- Semantic Commercial
analyst
ArcGIS Spatial Analyst ESRI http://desktop.arcgis.com/en/arcmap/10.3/tools/spatial-analyst-toolbox/ Region-based Commercial
an-overview-of-the-segmentation-and-classification-tools.htm
GeoSegment Chen (2018) http://130.15.95.215/lagisa/ Region-based Online tool, available
upon registration

based methods. However, finding the appropriate parameters for re- representation. Due to the size and nature (artificial entity) of pixels, in
gion-based methods is a significant challenge. Other drawbacks of re- many cases they do match with the size of the object. It would be ef-
gion-based methods are that they are complicated and time-consuming ficient if segmentation start from the perceptual meaningful entities
(Verma et al., 2011). On the other hand, HMs generate better results (Csillik, 2017). The idea of superpixels (which is low-level grouping of
compared to the edge- and region-based techniques, can utilize both the pixels) came from this concept. In contrast to pixels, superpixels carry
local and global homogeneity criterion, make the seed selection effi- more information and follow the natural image boundary (Zhengqin
cient, and can eliminate the noise effect. Even so, those algorithms are and Jiansheng, 2015). In addition, superpixels have an intermediate
difficult to implement, computation intensive, and no available soft- scale between pixels and objects (Achanta et al., 2012), reduce noise
ware on the market to execute. and outliers (Shi and Wang, 2014), and can speed up the subsequent
process. Achanta et al. (2012) categorized superpixels algorithms into
4.2. Key challenges for object-based segmentation methods graph and gradient ascent methods. Multiple algorithms have been
proposed to generate superpixels such as superpixels lattice (Moore
4.2.1. Segmentation of linear objects et al., 2008), turbopixels (Levinshtein et al., 2009), quick shift (Vedaldi
The techniques for segmenting different geographical objects can and Soatto, 2008) and simple linear iterative clustering (SLIC) (Achanta
vary substantially due to their physical and geometrical characteristics. et al., 2012). Stefanski et al. (2013) tested the performance of super-
For instance, spectral values of roads, the roof of buildings and turbid pixels contour in remote sensing image segmentation to optimize seg-
water are quite similar; however, the geometrical characteristics such mentation parameters and improve classification accuracy. They re-
as length-width ratio or linearity index vary widely among them. Road ported that it is easy to handle as only two parameters to deal with and
networks have enormous usage in many applications and a vital com- can optimize parameters selection process. Csillik (2017) proposed a
ponent in GIS systems. Based on the importance, a significant amount of segmentation workflow where famous MRS algorithm started from
research (Maboudi et al., 2016b; Mokhtarzade and Zoej, 2007; Shi superpixels instead of individual pixel. They stated that the proposed
et al., 2014; Sun and Messinger, 2013; P. Wang et al. (2015a,b,c) de- workflow significantly reduced processing time and provided better
voted especially on segmenting and extracting roads from high-re- accuracy.
solution images. By utilizing the spectral and geometric characteristics, Instead of using a single scale for the entire image, Fonseca-Luengo
directional segmentation (Chaudhuri et al., 2012), orientation-based et al. (2014) offered a hierarchical multiscale segmentation using su-
segmentation (Poullis and You, 2010), factorization-based segmenta- perpixels (SLIC) which allowed users to detect objects at different
tion (Yuan and Cheriyadat, 2013) have been proposed in the literature. scales. It can provide a realistic local optical scale (Gonzalo-Martín
Nevertheless, a reliable automated segmentation method for roads is et al., 2016) and a better understanding of land cover and its objects.
still far-off due to the nuisance caused by shadows, vegetation, sur- Yin and Yang (2017) compared superpixels with sub-pixels to map
rounding buildings, and other features on roads. urban green space and concluded that superpixels provided higher ac-
curacy than sub-pixels. Apart from above researches, recent trends are
4.2.2. Segmentation from low-level pixel grouping to incorporate superpixels with other models such as probabilistic fu-
Most segmentation algorithms use the pixel-grid as the underlying sion model (Zhang et al., 2015d), probability density function (Liu

124
 Table 5
Summary of different segmentation parameters optimization methods.
Optimization Algorithm/ Segmentation Cost Function/ Performance and Advantages Limitations Test data (sensor, spatial Examples
Approach Algorithm Criteria resolution, bands) and
application

Genetic Algorithm (GA) for MRS Area, number of accuracy improved only a single scale. Saba et al. (2016)
M.D. Hossain and D. Chen

optimizing segmentation segments and weight

• Classification
around 5% if compared to support
• Consider
proposed method can be
• IK1 m Others:
parameters coefficients. vector machine.
• The
tested by segmenting other
• No description for spectral Nikfar et al. (2012)
rid of manually selecting man-made features.
• bands and image size
• Get
parameters. road
Hybrid metaheuristics approach MRS Precision and recall seven strategies (single, The hybrid method requires
• Target:
QB, WV-2, aerial photograph Quirita et al. (2016)
for parameter tuning
• Compared
population and hybrid longer processing time compared
• m, 0.5 m, and 0.1 Others:
metaheuristics) regarding speed and to differential evolution.
• 2.44
respectively Happ et al. (2012)
proper solution. • RGB
• Target: building, swimming
pool, tank, and boats
An adaptive approach for scale MRS Thematic maps • Utilized the inherent features of • Many parameters were • GF-1 and ZY-3 Zhou et al. (2017a,b)
selection segmented objects and prior selected based on trial and • 16 m and 2.1 m for GF-1 and Others:
knowledge to improve segmentation error basis. ZY-3 respectively Zhou et al. (2017a), Qiu et al. (2016),
performance. • RGB + NIR Auquilla et al. (2014), Zhang et al.
• Target: built-up area, road, (2017a,b,c), Guo and Du (2017)
river, reservoir, forest,
farmland, wetland, and
grassland.
Spectral and spatial statistics for Mean Shift Local variance (LV) • Pre-estimated the optimal scale • Implemented only for • IK and QB Ming et al. (2015)
scale selection before segmentation from the panchromatic images. • 1 m and 0.7 m respectively Others with LV:
spatial statistics. • Required high computational • Panchromatic Drǎguţ et al. (2014, 2011, 2010), Zhao et al.
• Guaranteed high homogeneity and resources. • Target: buildings and (2012), Kavzoglu and Erdemir (2016),

125
high heterogeneity within and • The ideal scale does not exist farmland Grybas et al. (2017), Martha et al. (2011),
between segments. in an image with a nested Espindola et al. (2006)
• The method was independent of the structure.
spatial resolution of the image.
Spectral measures for scale MRS SA • Utilized all spectral bands. • Did not consider intra- • QB and WV-2 Yang et al. (2014)
selection • Identified multiple appropriate segment homogeneity and • 0.61 m and 0.5 m Others:
scales for different land cover within inter-segment heterogeneity. respectively Chubey et al. (2006)
the image. • Considered only the mean SA • Four-band pan-sharpen
which is not sensitive to the multispectral
heterogeneous image. • Target: buildings, vegetation,
impervious surfaces, etc.
Regression tree model for MRS Meta-analysis • Narrow down the range of suitable • Could not identify exact scale • Airborne, WV-2, and IK Johnson and Jozdani (2018)
generalizable scale scale parameters. for different land use. • 25 cm, 30 cm, 65 cm and
parameters • Considered a radiometric resolution 75 cm for airborne; 50 cm for
of an image. WV-2 and 1 m for IK
• RGB buildings, vegetation,
• Target:
road, bare soil, and water
Classification driven approach MRS GLCM • Identified best segmentation scale • Target was only vegetation • UAV Laliberte and Rango (2009)
for scale selection for sub-decimeter resolution UAV and bare land. cm Others:
images.
• 5RGB Dronova et al., (2012), Stumpf and Kerle
• Target: shrubs, grass, and (2011), Kalantar et al. (2017), Hadavand
• bare ground et al. (2017), Li and Shao (2013), Nichol and
Wong (2008), Peña-Barragán et al. (2011),
Li et al. (2014), Li and Shao (2014), Juel
et al. (2015)
Spatial autocorrelation for scale MRS Rate of Change • The global score method excluded • Identifying optimal scale from • Sensor: not specified Meng et al. (2014)
selection (ROC) of Moran’s I under-segmented scale. ROC-MI curve still • 0.08 m and 0.15 m Others:
(MI) • MI considered spatial distribution of challenging. • Target: urban features Johnson and Xie (2011), Johnson et al.
segments. (2015)
ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

et al., 2017), adaptive region merging (Ko and Ding, 2016), purpose even for a single image is a significant challenge (Chen et al., 2018a).
dependent grouping (Maboudi et al., 2016a), multiscale and multi-
feature normalized cut (Zhong et al., 2016), minimum spanning tree 4.2.5. Evaluation of segmentation results
(Wang et al., 2017c) and binary merge tree (Su and Zhang, 2018) to Optimal segmentation parameters selections methods intended to
identify optimal scale and parameters as well as to minimize under- and select parameters by post evaluation. Parameters are selected based on
over-segmentation problem. supervised and unsupervised method. Supervised method select para-
meters based on the similarity between the corresponding trial-and-
4.2.3. Multiscale segmentation error results and the reference data (Ghosh and Joshi, 2014; Wang
In high-resolution images, an individual object is modeled by many et al., 2018a,b). Similarity can be based on area overlap (Clinton et al.,
pixels. Pixels within an individual object tend to display high spectral 2010; Yang et al., 2015a; Zhang et al., 2015a), correctly matched ob-
autocorrelation. Even so, image objects exhibits an intrinsic scale, jects numbers (Liu et al., 2012; Marpu et al., 2010), object location
hierarchical structure and are composed of structurally associated (Montaghi et al., 2013), spectral discrepancy (Anders et al., 2011),
parts. As a result, Modifiable Areal Unit Problem (MAUP) (Marceau, border fitness (Albrecht, 2010; Neubert et al., 2008), or combination of
1999) is frequent in remote sensing images. Image objects can be these (Witharana and Civco, 2014; Zhang et al., 2015b). Unsupervised
treated differently in different scales. Therefore, multiscale segmenta- methods compare the resultant segments with the good segmentation
tion is an important issue for GEOBIA as a single scale is not suitable to (based on intra-segment homogeneity and inter-segment heterogeneity)
represent different image objects. However, there is no single optimal (Drǎguţ et al., 2014; Zhang et al., 2017b) and used in remote sensing as
scale (Hay et al., 2003), thus scholars are trying to identify scales that there is no true ground-truth segmentation of an image against which
are specific to the dominant image objects within a scene. the output of an algorithm can be compared. Both methods are facing
Among the segmentation algorithms, MRS has been widely used in difficulty in either generating reference objects or defining criteria that
the literature. The primary challenge in MRS is selecting appropriate can quantify intra-segment homogeneity and inter-segment hetero-
parameters as geographical objects varied in size, shape, and texture geneity between objects.
(Ma et al., 2015; Teodoro and Araujo, 2016). Among the parameters, Shapes of objects extracted from the segmentation are used by
scale plays a vital role in MRS. Selection of object-based scale in seg- classification algorithms to extract patterns for object labeling. They are
mentation is the key to GEOBIA because a wrong scale will lead to also used to assist in quantifying spectral statistics of each object. Apart
either over- or under-segmentation (Ming et al., 2012). In order to from the shape, the location of objects is also essential for geospatial
determine optimal scale, a trial-and-error method is commonly exe- analysis. Traditional pixel-based accuracy assessment methods are in-
cuted in remote sensing (Eisank et al., 2014; Ninsawat and Hossain, capable of calculating object shape and location (Clinton et al., 2010).
2016; Radoux and Defourny, 2007; Zhang et al., 2014c). However, the Several area- and shape-based goodness measures have been proposed
trial-and-error method is time-consuming and impractical for many in the literature (Clinton et al., 2010; Zhang, 1996) to judge segmen-
applications (Im et al., 2014; Ma et al., 2017). Multi-scale segmentation tation results. However, those measures use predefined objects from an
algorithms utilize user-defined scale in different ways. For instance, in image as the training objects. This process makes the segmentation
the FNEA scale determines the average size of the object whereas in WT accuracy assessment method somehow subjective as the training objects
it defines the sampling window size, valley and catchment area depend on human judgment.
threshold (Ming et al., 2015). Finding optimal scale is troublesome due
to several issues such as the implicit relationship between scale and 4.2.6. Image-objects vs geo-objects
image data, the intricate link between segmentation results on different Image-objects do not exist independently within digital images.
scales (Ming et al., 2012). Segmentation is the primary unit of GEOBIA and aims to identify
image-objects based on discreteness, coherency, and contrast (Castilla
4.2.4. Optimization of segmentation parameters and Hay, 2008). By contrast, a geographic object (geo-object) refers to
Apart from scale, texture also can increase segmentation accuracy an object having certain minimum size on or near the earth surface,
(Kim et al., 2011). Parameters optimization is a topic of research for with many permanent properties and differs from its surroundings
decades, and the recent trend is to employ automatic, optimal para- based on specific properties. When segmentation algorithms can gen-
meter determination procedure (Chen et al., 2018a). Optimal para- erate geo-objects, then segments are termed as meaningful image-ob-
meters will enhance intra-segment homogeneity, inter-segment het- jects. However, achieving meaningful image-objects is challenging due
erogeneity (Yang et al., 2015b), and classification accuracy (Gao et al., to complicated radiometric and semantic relationship between image-
2011). A collaborative approach (integration of thematic maps gener- and geo-objects and hierarchical details of objects. Thus, human in-
ated from the classification method to MRS) was implemented by terpretation of meaningful image-objects varies from segmentation al-
Troya-Galvis et al. (2016) to develop a generic segmentation procedure. gorithms. This conceptual gap is termed as over- and under-segmen-
Saba et al. (2016) introduced an automatic image segmentation method tation in segmentation results.
by using genetic algorithm optimization with a new cost function.
Furthermore, Esch et al. (2008) utilized fuzzy logic and iterative opti- 4.3. Future directions
mization respectively to identify optimal parameters. As indicated in
Table 5, multiple approaches have been proposed in the literature to Segmentation is the key component of GEOBIA by reducing image
identify optimal parameters for MRS and mean shift. In addition to complexity, making image content understandable and producing
those, some scholars have integrated MRS with other models. For ex- meaningful image objects (Lang, 2008). Pixels are the basic unit of a
ample, Li et al. (2008) proposed MRS by using Statistical Region Mer- raster image and usually square shaped. As pixels are not natural en-
ging (SRM) and Minimum Heterogeneity Rule (MHR). SRM was utilized tities, they do not match with the image content. By contrast, the
for initial segmentation and MHR for merging objects. Similarly, Gu hexagon is eligible to represent earth surface more efficiently (Sahr
et al. (2018) integrated graph-based segmentation with MRS where et al., 2003). Based on this concept, Hofmann and Tiede (2014) pro-
initial segments were generated by using graph theory, and merging posed hexagonal cell-based MRS approach. For the testing purpose,
was done by FNEA. Yang et al. (2015b) introduced a new energy they utilized WV-2 images, and their target was segmenting a soccer
function to quantify the relationship between image objects and its stadium. This method provided better results when compared to the
neighbors. Chen et al. (2012b) prescribed a soft image segmentation square cell-based MRS especially in segmenting linear and round
model based on multiresolution and probability of pixel merging at the shaped features. Another way of dealing square blocks is a low-level
top level. Nevertheless, defining appropriate segmentation parameters grouping of pixels which is more natural and efficient to work with

126
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

(Neubert and Protzel, 2012). The idea of superpixels (Ren and Malik, and weighted LV was proposed by Wang et al. (2018a,b) for historic
2003) also comes from the concept of low-level pixel grouping. The land use/land cover (LULC) change database updating. In this case, the
same evolution trend is expected in the future. segmentation process was guided by historical LULC boundaries. By
Geo-objects refer to the spatial entities formed by numerous ele- contrast, Golinkoff (2013) used objects area to guide the segmentation
ments distributed within a geographic area. In high-spatial resolution process.
images, geo-objects are distributed in a number of pixels. Thus, an
object that appears homogeneous in one scale may become hetero- 5. Summary
geneous at another scale. Although a significant amount of research
devoted to identifying optimal scale for segmentation, still the question With the advancement of remote sensing technology, high spatial
of “what is the optimum (or range at least) segmentation scale for resolution remote sensing images have been used widely in different
different image-objects within a scene?” roaming in remote sensing fields due to their rich spectral and textural information. GEOBIA has
arena. The scale has a multi-dimensional nature (Malenovský et al., evolved to analyze those high-resolution images. As a critical compo-
2007), complex hierarchy, and variability (Wu and Li, 2009). Studies nent of GEOBIA process, image segmentation algorithm has been a
have utilized geographical variance (Moellering and Tobler, 1972), hotspot recently. Though many algorithms have been proposed, all al-
wavelet transform (Percival, 1995), local variance (Woodcock and gorithms have some pros and cons. For instance, edge-based algorithms
Strahler, 1987) for measuring spatial structure. However, Nijland et al. are easy to implement, but they are missing the contextual information.
(2009) identified that there is no spatial scale appropriate for identi- The region-based method generates better results compared to the
fying and analyzing various urban features. What’s more, different edge-based method, however, finding appropriate seeds and other
segmentation algorithms treat scale in a different way. We may expect parameters is the real challenge in that case. To resolve the seeding
the advancement in dealing with the segmentation scale for generating problems, superpixels algorithms are introduced in remote sensing
meaningful image objects would make rapid progress. image segmentation. Another recent trend is to execute a hybrid
In addition to scale, segmentation results vary based on homo- method, although those algorithms are complicated and no software
geneity or heterogeneity criteria. Many algorithms have utilized tex- package available in the market to implement. Researches still trying to
tural parameters to developed rules for homogeneity and heterogeneity. identify algorithms (with optimal parameters) for the segmentation
As a bit change of homogeneity or heterogeneity leads to different process which can accurately identify individual image objects.
segmentation results, Hay and Castilla (2008) termed segmentation as Segmentation influences classification accuracy. However, using
“an ill-posed problem.” Thus, a substantial amount of research dedi- optimal parameters for segmentation algorithms is not the only solution
cated to segmentation parameters optimization (as shown in Table 5). for achieving higher accuracy in OBIA. Different parameter combina-
Nevertheless, more advanced techniques focusing on developing a tions would lead to similar classification results. Recently some re-
methodology of parameters optimization that is applicable in any searches have collaborated segmentation with classification in high-
context is deemed necessary. In addition, creating an object-based resolution image analysis (Heumann, 2011; Wang and Aldred, 2011;
segmentation dataset similar to “The Berkeley Segmentation Dataset Csillik, 2017; Guo and Du 2017; Hadavand et al. 2017). In those cases,
and Benchmark” (available at https://www2.eecs.berkeley.edu/ the segmentation does not have to be perfect. The classification process
Research/Projects/CS/vision/bsds/; which is not specific to remote may include object generating steps which will assist in overcoming the
sensing images) has become essential to compare the performance of over-segmentation problem and building the complex objects.
different segmentation algorithms and parameters in different image
settings. Declaration of interest
Although the recent trend is applying the hybrid segmentation
methods, these algorithms are facing problems due to implementation None.
complexity, and lack of available software and tools. More hybrid
segmentation techniques along with the tools will evolve in future. Funding
Another issue in these cases is applying algorithms on large dataset due
to the computational power of existing computer hardware. Several This research is supported by Canada National Science and
studies have implemented tile-wise segmentation and showed pro- Engineering Reseach Council (NSERC) Discovery grant.
mising results. However, these methods have several questions to an-
swer, dealing objects in the tile edges, removing processing window References
artifacts and dynamic selection of parameters based on the geo-objects
in each tile are among the most prominent. Another interesting topic in Achanta, R., Shaji, A., Smith, K., Lucchi, A., Fua, P., Süsstrunk, S., 2012. SLIC superpixels
segmentation is the evaluation of segmentation results. Object-based compared to state-of-the-art superpixel methods. IEEE Trans. Pattern Anal. Mach.
Intell. 34, 2274–2281. https://doi.org/10.1109/TPAMI.2012.120.
supervised and unsupervised methods are facing problem in creating Adams, R., Bischof, L., 1994. Seeded region growing. IEEE Trans. Pattern Anal. Mach.
reference objects and defining criteria, more research will focus on this Intell. 16, 641–647. https://doi.org/10.1109/34.295913.
topic. Akçay, H.G., Aksoy, S., 2008. Automatic detection of geospatial objects using multiple
hierarchical segmentations. IEEE Trans. Geosci. Remote Sens. 46, 2097–2111.
GEOBIA method based on the idea of one-to-one mapping between https://doi.org/10.1109/TGRS.2008.916644.
segments and image objects. Thus, classification algorithms require Al-Hujazi, E.H., Sood, A.K., 1991. Integration of edge- and region-based techniques for
perfect segments to provide accurate prediction. Based on the above range image segmentation. In: Casasent, D.P. (Ed.), SPIE 1381, Intelligent Robots and
Computer Vision IX: Algorithms and Techniques. International Society for Optics and
discussions, it can conclude that it is challenging to find an algorithm Photonics, pp. 589–599. https://doi.org/10.1117/12.25190.
that can generate perfect segments. In addition, many objects are Albrecht, F., 2010. Uncertainty in image interpretation as reference for accuracy assess-
composed of non-homogeneous regions in high-resolution images such ment in object-based image analysis. In: Ninth International Symposium on Spatial
Accuracy Assessment in Natural Resources and Environmental Sciences, pp. 13–16.
as the roof of a house (often composed of light and dark regions) which
Alshehhi, R., Marpu, R.P., 2017. Hierarchical graph-based segmentation for extracting
not likely to be segmented together. To resolve this issue, Troya-Galvis road networks from high-resolution satellite images. ISPRS J. Photogramm. Remote
et al. (2018) proposed a method to modify initial segments based on the Sens. 126, 245–260. https://doi.org/10.1016/j.isprsjprs.2017.02.008.
trained classifier. Their work is an extension of Collaborative segmen- Anders, N.S., Seijmonsbergen, A.C., Bouten, W., 2011. Segmentation optimization and
stratified object-based analysis for semi-automated geomorphological mapping.
tation-classification (CoSC) approach (Troya-Galvis et al., 2016). Al- Remote Sens. Environ. 115, 2976–2985. https://doi.org/10.1016/J.RSE.2011.05.
though the proposed method is time intensive and requires sufficient 007.
training data, it showed promising results. Furthermore, a scale self- Andrey, P., Tarroux, P., 1998. Unsupervised segmentation of Markov random field
modeled textured images using selectionist relaxation. IEEE Trans. Pattern Anal.
adaptive segmentation method based on exponential sampling scale

127
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

Mach. Intell. 20, 252–262. https://doi.org/10.1109/34.667883. S., Hay, G.J. (Eds.), Object-Based Image Analysis. Springer, Berlin Heidelberg, Berlin,
Audebert, N., Le Saux, B., Lefèvre, S., 2016. Semantic segmentation of earth observation Heidelberg, pp. 91–110. https://doi.org/10.1007/978-3-540-77058-9_5.
data using multimodal and multi-scale deep networks. In: Lai, S., Lepetit, V., Nishino, Castilla, G., Hay, G.J., Ruiz-Gallardo, J.R., 2008. Size-constrained Region Merging
K., Sato, Y. (Eds.), Asian Conference on Computer Vision. Springer, Cham, pp. (SCRM): an automated delineation tool for assisted photointerpretation.
180–196. https://doi.org/10.1007/978-3-319-54181-5_12. Photogramm. Eng. Remote Sens. 74, 409–419. https://doi.org/10.14358/PERS.74.4.
Auquilla, A., Heremans, S., Vanegas, P., Van Orshoven, J., 2014. A procedure for semi- 409.
automatic segmentation in OBIA based on the maximization of a comparison index. Chaudhuri, D., Kushwaha, N.K., Samal, A., 2012. Semi-automated road detection from
In: International Conference on Computational Science and Its Applications. Springer, high resolution satellite images by directional morphological enhancement and
Cham, pp. 360–375. https://doi.org/10.1007/978-3-319-09144-0_25. segmentation techniques. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 5,
Baatz, M., Schäpe, A., 2000. Multiresolution segmentation: an optimization approach for 1538–1544. https://doi.org/10.1109/JSTARS.2012.2199085.
high quality multi-scale image segmentation. Angew. Geogr. in- Cheevasuvit, F., Maitre, H., Vidal-Madjar, D., 1986. A robust method for picture seg-
formationsverarbeitung XII 58, 12–23. mentation based on a split-and-merge procedure. Comput. Vision Graph. Image
Ballard, D.H., 1981. Generalizing the Hough transform to detect arbitrary shapes. Pattern Process. 34, 268–281. https://doi.org/10.1016/S0734-189X(86)80042-1.
Recognit. 13, 111–122. https://doi.org/10.1016/0031-3203(81)90009-1. Chen, B., Qiu, F., Wu, B., Du, H., 2015. Image segmentation based on constrained spectral
Banerjee, B., Varma, S., Buddhiraju, K.M., 2012. Satellite image segmentation: a novel variance difference and edge penalty. Remote Sens. 7, 5980–6004. https://doi.org/
adaptive mean-shift clustering based approach. In: Geoscience and Remote Sensing 10.3390/rs70505980.
Symposium (IGARSS). IEEE International, pp. 4319–4322. Chen, D., 2018. LAGISA Online GeoSegment Tool [WWW Document]. URL http://130.15.
Barrile, V., Bilotta, G., 2016. Fast extraction of roads for emergencies with segmentation 95.215/lagisa/ (accessed 9.25.18).
of satellite imagery. Procedia -Social Behav. Sci. 223, 903–908. https://doi.org/10. Chen, G., Weng, Q., Hay, G.J., He, Y., 2018a. Geographic Object-based Image Analysis
1016/j.sbspro.2016.05.313. (GEOBIA): Emerging trends and future opportunities. GIScience Remote Sens. 55,
Beaulieu, J.M., Goldberg, M., 1989. Hierarchy in picture segmentation: a stepwise opti- 159–182. https://doi.org/10.1080/15481603.2018.1426092.
mization approach. IEEE Trans. Pattern Anal. Mach. Intell. 11, 150–163. https://doi. Chen, J., Deng, M., Mei, X., Chen, T., Shao, Q., Hong, L., 2014. Optimal segmentation of a
org/10.1109/34.16711. high-resolution remote-sensing image guided by area and boundary. Int. J. Remote
Bellens, R., Gautama, S., Martinez-Fonte, L., Philips, W., Chan, J.C.W., Canters, F., 2008. Sens. 35, 6914–6939. https://doi.org/10.1080/01431161.2014.960617.
Improved classification of VHR images of urban areas using directional morpholo- Chen, J., Li, J., Pan, D., Zhu, Q., Mao, Z., 2012a. Edge-guided multiscale segmentation of
gical profiles. IEEE Trans. Geosci. Remote Sens. 46, 2803–2813. https://doi.org/10. satellite multispectral imagery. IEEE Trans. Geosci. Remote Sens. 50, 4513–4520.
1109/TGRS.2008.2000628. https://doi.org/10.1109/TGRS.2012.2194502.
Benz, U.C., Hofmann, P., Willhauck, G., Lingenfelder, I., Heynen, M., 2004. Multi-re- Chen, J., Pan, D., Mao, Z., 2009. Image-object detectable in multiscale analysis on high-
solution, object-oriented fuzzy analysis of remote sensing data for GIS-ready in- resolution remotely sensed imagery. Int. J. Remote Sens. 30, 3585–3602. https://doi.
formation. ISPRS J. Photogramm. Remote Sens. 58, 239–258. https://doi.org/10. org/10.1080/01431160802585348.
1016/j.isprsjprs.2003.10.002. Chen, K.-M., Chen, S.-Y., 2002. Color texture segmentation using feature distributions.
Beveridge, J.R., Griffith, J., Kohler, R.R., Hanson, A.R., Riseman, E.M., 1989. Segmenting Pattern Recognit. Lett. 23, 755–771. https://doi.org/10.1016/S0167-8655(01)
images using localized histograms and region merging. Int. J. Comput. Vis. 2, 00150-7.
311–347. https://doi.org/10.1007/BF00158168. Chen, P.C., Pavlidis, T., 1979. Segmentation by texture using a co-occurrence matrix and
Bins, L.S.A., Fonseca, L.M.G., Erthal, G.J., Ii, F.M., 1996. Satellite Imagery Segmentation : a split-and-merge algorithm. Comput. Graph. Image Process. 10, 172–182. https://
a region growing approach. In: Anais VIII Simposia Brasilerio de Sensoriamenta doi.org/10.1016/0146-664X(79)90049-2.
Remoto. Slavador, Brazil, pp. 677–680. Chen, R., Li, X., Li, J., 2018b. Object-based features for house detection from RGB high-
Blaschke, T., 2010. Object based image analysis for remote sensing. ISPRS J. resolution images. Remote Sens. 10, 1–24. https://doi.org/10.3390/rs10030451.
Photogramm. Remote Sens. 65, 2–16. https://doi.org/10.1016/j.isprsjprs.2009.06. Chen, S.-Y., Lin, W.-C., Chen, C.-T., 1991. Split-and-merge image segmentation based on
004. localized feature analysis and statistical tests. CVGIP Graph. Model. Image Process.
Blaschke, T., Charles, B., Pekkarinen, A., 2004. Remote sensing image analysis: including 53, 457–475. https://doi.org/10.1016/1049-9652(91)90030-N.
the spatial domain. In: de Jong, S.M., van der Meer, F.D. (Ed.), Remote Sensing Image Chen, S., Luo, J., Shen, Z., Hu, X., Gao, L., 2008. Segmentation of multi-spectral satellite
Analysis: Including the Spatial Domain. Springer Netherlands, pp. 211–236. https:// images based on watershed algorithm. In: 2008 International Symposium on
doi.org/10.1017/S0032247400010123. Knowledge Acquisition and Modeling, KAM 2008. IEEE, pp. 684–688. https://doi.
Blaschke, T., Hay, G.J., Kelly, M., Lang, S., Hofmann, P., Addink, E., Feitosa, R.Q., Van org/10.1109/KAM.2008.84.
Der Meer, F., Van Der Werff, H., Van Coillie, F., Tiede, D., 2014. Geographic Object- Chen, X., Chen, J., Yamaguchi, Y., 2012. Soft image segmentation model. In: Proc. Int.
Based Image Analysis - Towards a new paradigm. ISPRS J. Photogramm. Remote Conf. Comput. Vis. Remote Sensing, CVRS 2012, pp. 90–93. https://doi.org/10.
Sens. 87, 180–191. https://doi.org/10.1016/j.isprsjprs.2013.09.014. 1109/CVRS.2012.6421239.
Blaschke, T., Lang, S., Hay, G.J. (Eds.), 2008. Object-based image analysis: spatial con- Chen, Z., Zhao, Z., Gong, P., Zeng, B., 2006. A new process for the segmentation of high
cepts for knowledge-driven remote sensing applications. Springer Science & Business resolution remote sensing imagery. Int. J. Remote Sens. 27, 4991–5001. https://doi.
Media. org/10.1080/01431160600658131.
Böhner, J., Selige, T., Ringeler, A., 2006. Image segmentation using representativeness Cheng, H.D., Jiang, X.H., Sun, Y., Wang, J., 2001. Color image segmentation: Advances
analysis and region growing. In: Boehner, J., McCloy, K.R., Strobl, J. (Eds.), SAGA - and prospects. Pattern Recognit. 34, 2259–2281. https://doi.org/10.1016/S0031-
Analyses and Modelling Applications. Göttinger Geographische Abhandlungen, 3203(00)00149-7.
Göttingen, Germany, pp. 29–38. Chen, Chu-Song, Wu, Ja-Ling, Hung, Yi-Ping, 1999. Theoretical aspects of vertically in-
Bouman, C.A., Shapiro, M., 1994. A multiscale random field model for Bayesian image variant gray-level morphological operators and their application on adaptive signal
segmentation. IEEE Trans. Image Process. 3, 162–177. https://doi.org/10.1109/83. and image filtering. IEEE Trans. Signal Process. 47, 1049–1060. https://doi.org/10.
277898. 1109/78.752602.
Bunting, P., Clewley, D., Lucas, R.M., Gillingham, S., 2014. The Remote Sensing and GIS Chubey, M.S., Franklin, S.E., Wulder, M.A., 2006. Object-based analysis of Ikonos-2
Software Library (RSGISLib). Comput. Geosci. 62, 216–226. https://doi.org/10. imagery for extraction of forest inventory parameters. Photogramm. Eng. Remote
1016/J.CAGEO.2013.08.007. Sens. 72, 383–394. https://doi.org/10.14358/PERS.72.4.383.
Burnett, C., Blaschke, T., 2003. A multi-scale segmentation/object relationship modelling Clinton, N., Holt, A., Scarborough, J., Yan, L., Gong, P., 2010. Accuracy assessment
methodology for landscape analysis. Ecol. Modell. 168, 233–249. https://doi.org/10. measures for object-based image segmentation goodness. Photogramm. Eng. Remote
1016/S0304-3800(03)00139-X. Sens. 76, 289–299. https://doi.org/10.14358/PERS.76.3.289.
Byun, Y., Kim, D., Lee, J., Kim, Y., 2011. A framework for the segmentation of high- Comaniciu, D., Meer, P., 2002. Mean shift: A robust approach toward feature space
resolution satellite imagery using modified seeded-region growing and region mer- analysis. IEEE Trans. Pattern Anal. Mach. Intell. 24, 603–619. https://doi.org/10.
ging. Int. J. Remote Sens. 32, 4589–4609. https://doi.org/10.1080/01431161.2010. 1109/34.1000236.
489066. Costa, H., Foody, G.M., Boyd, D.S., 2018. Supervised methods of image segmentation
Câmara, G., Souza, R.C.M., Freitas, U.M., Garrido, J., 1996. Spring: Integrating remote accuracy assessment in land cover mapping. Remote Sens. Environ. 205, 338–351.
sensing and gis by object-oriented data modelling. Comput. Graph. 20, 395–403. https://doi.org/10.1016/j.rse.2017.11.024.
https://doi.org/10.1016/0097-8493(96)00008-8. Csillik, O., 2017. Fast segmentation and classification of very high resolution remote
Campbell, J.B., Wynne, R.H., 2011. Introduction to Remote Sensing, 5th ed. The Guilford sensing data using. Remote Sens. 9, 19. https://doi.org/10.3390/rs9030243.
press, New York. D’Elia, C., Poggi, G., Scarpa, G., 2003. A tree-structured Markov random field model for
Canny, J., 1987. A computational approach to edge detection. In: Readings in Computer Bayesian image segmentation. IEEE Trans. image Process. 12, 1259–1273. https://
Vision. Elsevier, pp. 184–203. https://doi.org/10.1016/B978-0-08-051581-6. doi.org/10.1109/TIP.2003.817257.
50024-6. Davis, C.H., Wang, X., 2003. Planimetric accuracy of Ikonos 1 m panchromatic ortho-
Cánovas-García, F., Alonso-Sarría, F., 2015. A local approach to optimize the scale image products and their utility for local government GIS basemap applications. Int.
parameter in multiresolution segmentation for multispectral imagery. Geocarto Int. J. Remote Sens. 24, 4267–4288. https://doi.org/10.1080/0143116031000070328.
30, 937–961. https://doi.org/10.1080/10106049.2015.1004131. Davis, L.S., Rosenfeld, A., Weszka, J.S., 1975. Region extraction by averaging and
Cao, W., Li, J., Liu, J., Zhang, P., 2016. Two improved segmentation algorithms for whole thresholding. IEEE Trans. Syst. Man. Cybern. SMC-5, 383–388. https://doi.org/10.
cardiac CT sequence images. In: International Congress on Image and Signal 1109/TSMC.1975.5408419.
Processing, BioMedical Engineering and Informatics (CISP-BMEI). IEEE, pp. 346–351. De Smet, P., Pries, R.L.V.P.M., 2000. Implementation and analysis of an optimized
Carleer, A.P., Debeir, O., Wolff, E., 2005. Assessment of very high spatial resolution sa- rainfalling watershed algorithm. Electron. Imaging 8, 759–1166.
tellite image segmentations. Photogramm. Eng. Remote Sens. 71, 1285–1294. Deng, F.L., Tang, P., Liu, Y., Yang, C.J., 2013. Automated hierarchical segmentation of
https://doi.org/10.14358/PERS.71.11.1285. high-resolution remote sensing imagery with introduced relaxation factors. J. Remote
Castilla, G., Hay, G.J., 2008. Image objects and geographic objects. In: Blaschke, T., Lang, Sens. 17, 1492–1507.

128
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

Deng, Y., Manjunath, B.S., 2001. Unsupervised segmentation of color-texture regions in Geosci. Remote Sens. 53, 2987–3004. https://doi.org/10.1109/TGRS.2014.2367129.
images and video. IEEE Trans. Pattern Anal. Mach. Intell. 23, 800–810. https://doi. Gaetano, R., Masi, G., Scarpa, G., Poggi, G., 2012. A marker-controlled watershed seg-
org/10.1109/34.946985. mentation: Edge, mark and fill. In: International Geoscience and Remote Sensing
Deriche, M., Amin, A., Qureshi, M., 2017. Color image segmentation by combining the Symposium (IGARSS). IEEE, Munich, Germany, pp. 4315–4318. https://doi.org/10.
convex active contour and the Chan Vese model. Pattern Anal. Appl. 1–15. https:// 1109/IGARSS.2012.6351713.
doi.org/10.1007/s10044-017-0632-9. Gaetano, R., Scarpa, G., Poggi, G., 2009. Hierarchical texture-based segmentation of
Deriche, R., 1990. Fast algorithms for low-level vision. IEEE Trans. Pattern Anal. Mach. multiresolution remote-sensing images. IEEE Trans. Geosci. Remote Sens. 47,
Intell. 12, 78–87. https://doi.org/10.1109/34.41386. 2129–2141. https://doi.org/10.1109/TGRS.2008.2010708.
Derivaux, S., Forestier, G., Wemmert, C., Lefvre, S., 2010. Supervised image segmentation Gambotto, J.-P., 1993. A new approach to combining region growing and edge detection.
using watershed transform, fuzzy classification and evolutionary computation. Pattern Recognit. Lett. 14, 869–875.
Pattern Recognit. Lett. 31, 2364–2374. https://doi.org/10.1016/j.patrec.2010.07. Gao, H., Tang, Y., Jing, L., Li, H., Ding, H., 2017. A novel unsupervised segmentation
007. quality evaluation method for remote sensing images. Sensors 17, 2427. https://doi.
Dey, V., Zhang, Y., Zhong, M., 2010. A review on image segmentation techniques with org/10.3390/s17102427.
remote sensing perspective. In: Wagner, W., Székely, B. (Ed.), ISPRS TC VII Gao, Y.A.N., Mas, J.F., Kerle, N., Navarrete Pacheco, J.A., 2011. Optimal region growing
Symposium – 100 Years ISPRS. Vienna, pp. 31–42. segmentation and its effect on classification accuracy. Int. J. Remote Sens. 32,
Dey, V., Zhang, Y., Zhong, M., Salehi, B., 2013. Image segmentation techniques for urban 3747–3763. https://doi.org/10.1080/01431161003777189.
land cover segmentation of VHR imagery: Recent developments and future prospects. Geman, S., Graffigne, C., 1986. Markov random field image models and their applications
Int. J. Geoinform. 9, 15–35. to computer vision. In: International Congress of Mathematicians, pp. 1496–1517.
Drǎguţ, L., Csillik, O., Eisank, C., Tiede, D., 2014. Automated parameterisation for multi- Ghita, O., Whelan, P.F., 2002. Computational approach for edge linking. J. Electron.
scale image segmentation on multiple layers. ISPRS J. Photogramm. Remote Sens. 88, Imaging 11, 479. https://doi.org/10.1117/1.1501574.
119–127. https://doi.org/10.1016/j.isprsjprs.2013.11.018. Ghosh, A., Joshi, P.K., 2014. A comparison of selected classification algorithms for
Drǎguţ, L., Eisank, C., Strasser, T., 2011. Local variance for multi-scale analysis in geo- mappingbamboo patches in lower Gangetic plains using very high resolution
morphometry. Geomorphology 130, 162–172. https://doi.org/10.1016/j.geomorph. WorldView 2 imagery. Int. J. Appl. Earth Obs. Geoinf. 26, 298–311. https://doi.org/
2011.03.011. 10.1016/j.jag.2013.08.011.
Drǎguţ, L., Tiede, D., Levick, S.R., 2010. ESP: a tool to estimate scale parameter for Gofman, E., 2006. Developing an efficient region growing engine for image segmentation.
multiresolution image segmentation of remotely sensed data. Int. J. Geogr. Inf. Sci. In: Proceedings - International Conference on Image Processing. ICIP, pp. 2413–2416.
24, 859–871. https://doi.org/10.1080/13658810903174803. https://doi.org/10.1109/ICIP.2006.312949.
Dronova, I., Gong, P., Clinton, N.E., Wang, L., Fu, W., Qi, S., Liu, Y., 2012. Landscape Golinkoff, J.S., 2013. Area dependent region merging: A novel, user-customizable method
analysis of wetland plant functional types: The effects of image segmentation scale, to create forest stands and strata. Eur. J. Remote Sens. 46, 511–533. https://doi.org/
vegetation classes and classification methods. Remote Sens. Environ. 127, 357–369. 10.5721/EuJRS20134630.
https://doi.org/10.1016/j.rse.2012.09.018. Gonzalo-Martín, C., Lillo-Saavedra, M., Menasalvas, E., Fonseca-Luengo, D., García-
Du, S., Guo, Z., Wang, W., Guo, L., Nie, J., 2016. A comparative study of the segmentation Pedrero, A., Costumero, R., 2016. Local optimal scale in a hierarchical segmentation
of weighted aggregation and multiresolution segmentation. GIScience Remote Sens. method for satellite images: An OBIA approach for the agricultural landscape. J.
53, 1–20. https://doi.org/10.1080/15481603.2016.1215769. Intell. Inf. Syst. 46, 517–529. https://doi.org/10.1007/s10844-015-0365-4.
Ďurikovič, R., Kaneda, K., Yamashita, H., 1995. Dynamic contour: A texture approach and Grizonnet, M., Michel, J., Poughon, V., Inglada, J., Savinaud, M., Cresson, R., 2017. Orfeo
contour operations. Vis. Comput. 11, 277–289. https://doi.org/10.1007/ ToolBox: open source processing of remote sensing images. Open Geospatial Data,
BF01898405. Softw. Stand. 2, 15. https://doi.org/10.1186/s40965-017-0031-6.
Eisank, C., Smith, M., Hillier, J., 2014. Assessment of multiresolution segmentation for Grybas, H., Melendy, L., Congalton, R.G., 2017. A comparison of unsupervised segmen-
delimiting drumlins in digital elevation models. Geomorphology 214, 452–464. tation parameter optimization approaches using moderate- and high-resolution
https://doi.org/10.1016/j.geomorph.2014.02.028. imagery. GIScience Remote Sens. 54, 515–533. https://doi.org/10.1080/15481603.
Epshtein, B., Ofek, E., Wexler, Y., 2010. Detecting text in natural scenes with stroke width 2017.1287238.
transform. In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Gu, H., Han, Y., Yang, Y., Li, H., Liu, Z., Soergel, U., Blaschke, T., Cui, S., 2018. An
IEEE, pp. 2963–2970. https://doi.org/10.1109/CVPR.2010.5540041. efficient parallel multi-scale segmentation method for remote sensing imagery.
Esch, T., Thiel, M., Bock, M., Roth, A., Dech, S., 2008. Improvement of image segmen- Remote Sens. 10, 590. https://doi.org/10.3390/rs10040590.
tation accuracy based on multiscale optimization procedure. IEEE Geosci. Remote Guindon, B., 1997. Computer-based aerial image understanding: A review and assessment
Sens. Lett. 5, 463–467. https://doi.org/10.1109/LGRS.2008.919622. of its application to planimetric information extraction from very high resolution
Espindola, G.M., Camara, G., Reis, I.a., Bins, L.S., Monteiro, a.M., 2006. Parameter se- satellite images. Can. J. Remote Sens. 23, 38–47. https://doi.org/10.1080/
lection for region-growing image segmentation algorithms using spatial auto- 07038992.1997.10874676.
correlation. Int. J. Remote Sens. 27, 3035–3040. https://doi.org/10.1080/ Guo, Z., Du, S., 2017. Mining parameter information for building extraction and change
01431160600617194. detection with very high-resolution imagery and GIS data. GISci. Remote Sens. 54,
Falah, R.K., Ph.Bolon, J.P.Cocquerez, 1994. A region-region and region-edge cooperative 38–63. https://doi.org/10.1080/15481603.2016.1250328.
approach of image segmentation. In: International Conference on Image Processing. Hadavand, A., Saadatseresht, M., Homayouni, S., 2017. Segmentation parameter selection
IEEE, Austin, Texas, Texas, pp. 470–474. for object-based land-cover mapping from ultra high resolution spectral and elevation
Fan, J., Yau, D.K., Elmagarmid, A.K., Aref, W.G., 2001. Automatic image segmentation by data. Int. J. Remote Sens. 38, 3586–3607. https://doi.org/10.1080/01431161.2017.
integrating color-edge extractionand seeded region growing. IEEE Trans. Image 1302107.
Process. 10, 1454–1466. https://doi.org/10.1109/83.951532. Happ, P.N., Feitosa, R.Q., Street, A., 2012. Assessment of optimization methods for au-
Fan, J., Zeng, G., Body, M., Hacid, M.S., 2005. Seeded region growing: An extensive and tomatic tuning of segmentation parameters. In: 4th International Conference on
comparative study. Pattern Recognit. Lett. 26, 1139–1156. https://doi.org/10.1016/ Geographic Object-Based Image Analysis. Rio de Janeiro, pp. 490–495.
j.patrec.2004.10.010. Happ, P.N., Ferreira, R.S., Costa, G.A.O.P., Feitosa, R.Q., Bentes, C., Farias, R.,
Farag, A.A., Mohamed, R.M., El-Baz, A., 2005. A unified framework for MAP estimation in Achanccaray, P.M., 2016. Interseg: a distributed image segmentation tool. In:
remote sensing image segmentation. IEEE Trans. Geosci. Remote Sens. 43, GEOBIA 2016: Solutions and Synergies. University of Twente Faculty of Geo-
1617–1634. https://doi.org/10.1109/TGRS.2005.849059. Information and Earth Observation (ITC). https://doi.org/10.3990/2.450.
Farid, H., Simoncelli, E.P., 1997. Optimally rotation-equivariant directional derivative Haralick, R.M., 1981. Edge and region analysis for digital image data. In: Image
kernels. In: International Conference on Computer Analysis of Images and Patterns. Modeling. Elsevier, pp. 171–184. https://doi.org/10.1016/B978-0-12-597320-5.
Springer, Berlin, Heidelberg, pp. 207–214. https://doi.org/10.1007/3-540-63460-6_ 50014-0.
119. Haralick, R.M., Shapiro, L.G., 1985. Image Segmentation Techniques. Comput. Vision
Felzenszwalb, P.F., Huttenlocher, D.P., 2004. Efficient graph-based image segmentation. Graph. Image Process. 29, 100–132.
Int. J. Comput. Vis. 59, 167–181. https://doi.org/10.1023/B:VISI.0000022288. Haris, K., Efstratiadis, S.N., Maglaveras, N., Katsaggelos, A.K., 1998. Hybrid image seg-
19776.77. mentation using watersheds and fast region merging. IEEE Trans. Image Process. 7,
Feng, W., Jia, J., Liu, Z.Q., 2010. Self-validated labeling of Markov random fields for 1684–1699. https://doi.org/10.1109/83.730380.
image segmentation. IEEE Trans. Pattern Anal. Mach. Intell. 32, 1871–1887. https:// Hay, G.J., Blaschke, T., Marceau, D.J., Bouchard, A., 2003. A comparison of three image-
doi.org/10.1109/TPAMI.2010.24. object methods for the multiscale analysis of landscape structure. ISPRS J.
Fisher, P., 1997. The pixel: a snare and a delusion. Int. J. Remote Sens. 18 (3), 679–685. Photogramm. Remote Sens. 57, 327–345. https://doi.org/10.1016/S0924-2716(02)
Fonseca-Luengo, D., García-Pedrero, A., Lillo-Saavedra, M., Costumero, R., Menasalvas, 00162-4.
E., Gonzalo-Martín, C., 2014. Optimal scale in a hierarchical segmentation method Hay, G.J., Castilla, G., 2008. Chapter 1.4 Geographic Object-Based Image Analysis
for satellite images. In: International Conference on Rough Sets and Intelligent (GEOBIA): A new name for a new discipline. In: Blaschke, T., Lang, S., Hay, G.J.
Systems Paradigms. Springer, pp. 351–358. https://doi.org/10.1007/978-3-319- (Eds.), Object-Based Image Analysis: Spatial Concepts for Knowledge-Driven Remote
08729-0_36. Sensing Applications. Springer Science & Business Media, pp. 75–90.
Fosgate, C.H., Krim, H., Irving, W.W., Karl, W.C., Willsky, A.S., 1997. Multiscale seg- Hay, G.J., Castilla, G., 2006. Object-Based Image Analysis: Strengths, Weaknesses,
mentation and anomaly enhancement of SAR imagery. IEEE Trans. Image Process. 6, Opportunities and Threats (SWOT). In: 1st International Conference on Object-Based
7–20. https://doi.org/10.1109/83.552077. Image Analysis (OBIA 2006), pp. 4–5. https://doi.org/10.1007/978-3-540-77058-
Freixenet, J., Muñoz, X., Raba, D., Martí, J., Cufí, X., 2002. Yet another survey on image 9_4.
segmentation: Region and boundary information integration. Comput. Vision—ECCV Hay, G.J., Castilla, G., Wulder, M.A., Ruiz, J.R., 2005. An automated object-based ap-
2002, pp. 21–25. proach for the multiscale image segmentation of forest scenes. Int. J. Appl. Earth Obs.
Gaetano, R., Masi, G., Poggi, G., Verdoliva, L., Scarpa, G., 2015. Marker-controlled wa- Geoinf. 7, 339–359. https://doi.org/10.1016/j.jag.2005.06.005.
tershed-based segmentation of multiresolution remote sensing images. IEEE Trans. He, J., Kim, C., Kuo, C.-C.J., 2014. Interactive Segmentation Techniques Algorithms and

129
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

Performance Evaluation. Springer. Kelkar, D., Gupta, S., 2008. Improved Quadtree Method for Split Merge Image
Heumann, B.W., 2011. An object-based classification of mangroves using a hybrid deci- Segmentation. In: 2008 First International Conference on Emerging Trends in
sion tree-support vector machine approach. Remote Sens. 3, 2440–2460. https://doi. Engineering and TechnologyE. IEE, pp. 44–47. https://doi.org/10.1109/ICETET.
org/10.3390/rs3112440. 2008.145.
Hofmann, P., Tiede, D., 2014. Image segmentation based on hexagonal sampling grids. Kemker, R., Salvaggio, C., Kanan, C., 2018. Algorithms for semantic segmentation of
South-Eastern Eur. J. Earth Obs. Geomat. 3, 173–177. multispectral remote sensing imagery using deep learning. ISPRS J. Photogramm.
Horowitz, S.L., Pavlidis, T., 1976. Picture segmentation by a tree traversal algorithm. J. Remote Sens. https://doi.org/10.1016/j.isprsjprs.2018.04.014.
ACM 23, 368–388. https://doi.org/10.1145/321941.321956. Kerem, S., Ulusoy, I., 2013. Automatic multi-scale segmentation of high spatial resolution
Hu, X., Tao, C.V., Prenzel, B., 2005. Automatic segmentation of high-resolution satellite satellite images using watersheds. In: Geoscience and Remote Sensing Symposium
imagery by integrating texture, intensity, and color features. Photogramm. Eng. (IGARSS). IEEE International, pp. 2505–2508.
Remote Sens. 71, 1399. https://doi.org/10.14358/PERS.71.12.1399. Kerfoot, I.B., Bresler, Y., 1999. Theoretical analysis of multispectral image segmentation
Hu, Y., Chen, J., Pan, D., Hao, Z., 2016. Edge-guided image object detection in multiscale criteria. IEEE Trans. image Process. 8, 798–820. https://doi.org/10.1109/83.766858.
segmentation for high-resolution remotely sensed imagery. IEEE Trans. Geosci. Kermad, C.D., Chehdi, K., 2002. Automatic image segmentation system through iterative
Remote Sens. 54, 4702–4711. https://doi.org/10.1109/TGRS.2016.2550059. edge-region co-operation. Image Vis. Comput. 20, 541–555. https://doi.org/10.
Huang, Z., Zhang, J., Li, X., Zhang, H., 2014. Remote sensing image segmentation based 1016/S0262-8856(02)00043-4.
on Dynamic Statistical Region Merging. Opt. – Int. J. Light Electron Opt. 125, Kim, M., Warner, T.A., Madden, M., Atkinson, D.S., 2011. Multi-scale GEOBIA with very
870–875. https://doi.org/10.1016/j.ijleo.2013.07.092. high spatial resolution digital aerial imagery: scale, texture and image objects. Int. J.
Iannizzotto, G., Vita, L., 2000. Fast and accurate edge-based segmentation with no con- Remote Sens. 32, 2825–2850. https://doi.org/10.1080/01431161003745608.
tour smoothing in 2-D real images. IEEE Trans. Image Process. 9, 1232–1237. https:// Kiryati, N., Eldar, Y., 1991. A probabilistic hough transform. Pattern Recognit. 24,
doi.org/10.1109/83.847835. 303–316.
Ikonomopoulos, A., 1982. An approach to edge detection based on the direction of edge Ko, H.-Y., Ding, J.-J., 2016. Adaptive growing and merging algorithm for image seg-
elements. Comput. Graph. Image Process. 19, 179–195. https://doi.org/10.1016/ mentation. In: 2016 Asia-Pacific Signal and Information Processing Association
0146-664X(82)90107-1. Annual Summit and Conference (APSIPA). IEEE, Jeju, South Korea. https://doi.org/
Im, J., Quackenbush, L.J., Li, M., Fang, F., 2014. Optimum scale in object-based image 10.1109/APSIPA.2016.7820762.
analysis. In: Weng, Q. (Ed.), Scale Issues in Remote Sensing. John Wiley & Sons Ltd., Körting, T.S., Castejon, E.F., Fonseca, L.M.G., 2013. The divide and segment method for
New Jersey, pp. 197–214. parallel image segmentation. In: Advanced Concepts for Intelligent Vision Systems.
Jain, R., Kasturi, R., Schunck, B.G., 1995. Machine Vision. McGraw-Hill. Springer, pp. 504–515. https://doi.org/10.1007/978-3-319-02895-8_45.
Jaing, J.-A., Chuang, C.-L., Lu, Y.-L., Fahn, C.-S., 1994. Mathematical-morphology-based Krishnamachari, S., Chellappa, R., 1997. Multiresolution Gauss-Markov random field
edge detectors for detection of thin edges in low-contrast regions. IET Image Process. models for texture segmentation. IEEE Trans. Image Process. 6, 251–267. https://doi.
1, 269–277. org/10.1109/83.551696.
Jevtic, A., Melgar, I., Andina, D., 2009. Ant based edge linking algorithm. In: 35th Annual Kruse, F.A., Lefkoff, A.B., Boardman, J.W., Heidebrecht, K.B., Shapiro, A.T., Barloon, P.J.,
Conference of IEEE on Industrial Electronics, 2009. IEEE, pp. 3353–3358. https://doi. Goetz, A.F.H., 1993. The spectral image processing system (SIPS)—interactive vi-
org/10.1109/IECON.2009.5415195. sualization and analysis of imaging spectrometer data. Remote Sens. Environ. 44,
Jiao, L., Gong, M., Wang, S., Hou, B., Zheng, Z., Wu, Q., 2010. Natural and remote sensing 145–163. https://doi.org/10.1016/0034-4257(93)90013-N.
image segmentation using memetic computing. IEEE Comput. Intell. Mag. 5, 78–91. Kundu, M.K., Pal, S.K., 1986. Thresholding for edge detection using human psychovisual
https://doi.org/10.1109/MCI.2010.936307. phenomena. Pattern Recognit. Lett. 4, 433–441. https://doi.org/10.1016/0167-
Johnson, B., Bragais, M., Endo, I., Magcale-Macandog, D., Macandog, P., 2015. Image 8655(86)90041-3.
segmentation parameter optimization considering within- and between-segment Kurnaz, M.N., Dokur, Z., Ölmez, T., 2005. Segmentation of remote-sensing images by
heterogeneity at multiple scale levels: test case for mapping residential areas using incremental neural network. Pattern Recognit. Lett. 26, 1096–1104. https://doi.org/
landsat imagery. ISPRS Int. J. Geo-Information 4, 2292–2305. https://doi.org/10. 10.1016/J.PATREC.2004.10.004.
3390/ijgi4042292. Laliberte, A.S., Rango, A., 2009. Texture and scale in object-based analysis of sub-
Johnson, B., Jozdani, S., 2018. Identifying generalizable image segmentation parameters decimeter resolution unmanned aerial vehicle (UAV) imagery. IEEE Trans. Geosci.
for urban land cover mapping through meta-analysis and regression tree modeling. Remote Sens. 47, 1–10. https://doi.org/10.1109/TGRS.2008.2009355.
Remote Sens. 10, 73. https://doi.org/10.3390/rs10010073. Lang, S., 2008. Object-based image analysis for remote sensing applications : modeling
Johnson, B., Xie, Z., 2011. Unsupervised image segmentation evaluation and refinement reality – dealing with complexity. In: Object-Based Image Analysis. Berlin,
using a multi-scale approach. ISPRS J. Photogramm. Remote Sens. 66, 473–483. Heidelberg, pp. 3–27.
https://doi.org/10.1016/j.isprsjprs.2011.02.006. Levinshtein, A., Stere, A., Kutulakos, K.N., Fleet, D.J., Dickinson, S.J., Siddiqi, K., 2009.
Johnson, B.A., 2013. High-resolution urban land-cover classification using a competitive TurboPixels: Fast superpixels using geometric flows. IEEE Trans. Pattern Anal. Mach.
multi-scale object-based approach. Remote Sens. Lett. 4, 131–140. https://doi.org/ Intell. 31, 2290–2297. https://doi.org/10.1109/TPAMI.2009.96.
10.1080/2150704X.2012.705440. Levner, I., Zhang, H., 2007. Classification-driven watershed segmentation. IEEE Trans.
Judah, A., Hu, B., Wang, J., 2014. An algorithm for boundary adjustment toward multi- Image Process. 16, 1437–1445. https://doi.org/10.1109/TIP.2007.894239.
scale adaptive segmentation of remotely sensed imagery. Remote Sens. 6, Leymarie, F., Levine, M.D., 1993. Tracking deformable objects in the plane using an ac-
3583–3610. https://doi.org/10.3390/rs6053583. tive contour model. IEEE Trans. Pattern Anal. Mach. Intell. 15, 617–634. https://doi.
Juel, A., Groom, G.B., Svenning, J.-C., Ejrnæs, R., 2015. Spatial application of Random org/10.1109/34.216733.
Forest models for fine-scale coastal vegetation classification using object based Li, B., Pan, M., Wu, Z., 2012. An Improved Segmentation of High Spatial Resolution
analysis of aerial orthophoto and DEM data. Int. J. Appl. Earth Obs. Geoinf. 42, Remote Sensing Image using Marker-based Watershed Algorithm. In: Geoinformatics
106–114. https://doi.org/10.1016/J.JAG.2015.05.008. (GEOINFORMATICS), 2012 20th International Conference On. IEEE, pp. 1–5.
Jung, M., Yun, E.J., Kim, C.S., 2005. Multiresolution approach for texture segmentation Li, D., Zhang, G., Wu, Z., Yi, L., 2010a. An edge embedded marker-based watershed al-
using MRF models. In: International Geoscience and Remote Sensing Symposium gorithm for high spatial resolution remote sensing image segmentation. IEEE Trans.
(IGARSS). IEEE. https://doi.org/10.1109/IGARSS.2005.1525782. Image Process. 19, 2781–2787. https://doi.org/10.1109/TIP.2010.2049528.
Kaganami, H.G., Beiji, Z., 2009. Region-based segmentation versus edge detection. In: Li, H., Gu, H., Han, Y., Yang, J., 2010b. Object-oriented classification of high-resolution
2009 5th International Conference on Intelligent Information Hiding and Multimedia remote sensing imagery based on an improved colour structure code and a support
Signal Processing. IEEE, pp. 1217–1221. https://doi.org/10.1109/IIH-MSP.2009.13. vector machine. Int. J. Remote Sens. 31, 1453–1470. https://doi.org/10.1080/
Kalantar, B., Mansor, S. Bin, Sameen, M.I., Pradhan, B., Shafri, H.Z.M., 2017. Drone-based 01431160903475266.
land-cover mapping using a fuzzy unordered rule induction algorithm integrated into Li, H., Gu, H., Han, Y., Yang, J., 2009. An efficient multiscale SRMMHR (Statistical Region
object-based image analysis. Int. J. Remote Sens. 38, 2535–2556. https://doi.org/10. Merging and Minimum Heterogeneity Rule) segmentation method for high-resolution
1080/01431161.2016.1277043. remote sensing imagery. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2, 67–73.
Karl, J.W., Maurer, B.A., 2010. Spatial dependence of predictions from image segmen- https://doi.org/10.1109/JSTARS.2009.2022047.
tation: A variogram-based method to determine appropriate scales for producing Li, H.T., Gu, H.Y., Han, Y.S., Yang, J.S., 2008. An efficient multi-scale segmentation for
land-management information. Ecol. Inform. 5, 194–202. https://doi.org/10.1016/j. highresolution remote sensing imagery based on Statistical region merging and
ecoinf.2010.02.004. minimum heterogeneity rule. Int. Work. Earth Obs. Remote Sens. Appl 1257–1262.
Kass, M., Witkin, A., Terzopoulos, D., 1988. Snakes: Active contour models. Int. J. Li, M., Ma, L., Blaschke, T., Cheng, L., Tiede, D., 2016. A systematic comparison of dif-
Comput. Vis. 1, 321–331. https://doi.org/10.1007/BF00133570. ferent object-based classification techniques using high spatial resolution imagery in
Kaur, B., Garg, A., 2011. Mathematical morphological edge detection for remote sensing agricultural environments. Int. J. Appl. Earth Obs. Geoinf. 49, 87–98. https://doi.
images. In: ICECT 2011 - 2011 3rd Int. Conf. Electron. Comput. Technol., vol. 5, pp. org/10.1016/j.jag.2016.01.011.
324–327. https://doi.org/10.1109/ICECTECH.2011.5942012. Li, N., Huo, H., Fang, T., 2010c. A novel texture-preceded segmentation algorithm for hr
Kavzoglu, T., Erdemir, M.Y., 2016. A hierarchical scale setting strategy for improved image.pdf. IEEE Trans. Geosci. Remote Sens. 48, 2818–2828.
segmentation performance using very high resolution images. Spatial Accuracy. Li, P., Guo, J., Song, B., Xiao, X., 2011. A multilevel hierarchical image segmentation
195–201. method for urban impervious surface mapping using very high resolution imagery.
Kavzoglu, T., Tonbul, H., 2017. A comparative study of segmentation quality for multi- IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 4, 103–116. https://doi.org/10.
resolution segmentation and watershed transform. In: Proc. 8th Int. Conf. Recent 1109/JSTARS.2010.2074186.
Adv. Sp. Technol. RAST 2017, pp. 113–117. https://doi.org/10.1109/RAST.2017. Li, X., Myint, S.W., Zhang, Y., Galletti, C., Zhang, X., Turner, B.L., 2014. Object-based
8002984. land-cover classification for metropolitan Phoenix, Arizona, using aerial photo-
Kavzoglu, T., Yildiz, E.M., Tonbul, H., 2016. A region-based multi-scale approach for graphy. Int. J. Appl. Earth Obs. Geoinf. 33, 321–330. https://doi.org/10.1016/J.JAG.
object-based image analysis. ISPRS - Int. Arch. Photogramm. Remote Sens Spat. Inf. 2014.04.018.
Sci. 41. https://doi.org/10.5194/isprs-archives-XLI-B7-241-2016. Li, X., Shao, G., 2014. Object-based land-cover mapping with high resolution aerial

130
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

photography at a county scale in midwestern USA. Remote Sens. 6, 11372–11390. segmentation results. J. Spat. Sci. 55, 55–68. https://doi.org/10.1080/14498596.
https://doi.org/10.3390/rs61111372. 2010.487850.
Li, X., Shao, G., 2013. Object-based urban vegetation mapping with high-resolution aerial Marr, D., Hildreth, E., 1980. Theory of edge detection. Proc. R. Soc. London Ser. BBiol.
photography as a single data source. Int. J. Remote Sens. 34, 771–789. https://doi. Sci. 207, 187–217. https://doi.org/10.1098/RSPB.1980.0020.
org/10.1080/01431161.2012.714508. Martha, T.R., Kerle, N., Van Westen, C.J., Jetten, V., Kumar, K.V., 2011. Segment opti-
Lin, G., Adiga, U., Olson, K., Guzowski, J.F., Barnes, C.A., Roysam, B., 2003. A hybrid 3D mization and data-driven thresholding for knowledge-based landslide detection by
watershed algorithm incorporating gradient cues and object models for automatic object-based image analysis. IEEE Trans. Geosci. Remote Sens. 49, 4928–4943.
segmentation of nuclei in confocal image stacks. Cytometry 56, 23–36. https://doi. https://doi.org/10.1109/TGRS.2011.2151866.
org/10.1002/cyto.a.10079. Martin, D.R., Fowlkes, C.C., Malik, J., 2004. Learning to detect natural image boundaries
Ling, F., Li, X., Xiao, F., Fang, S., Dub, Y., 2012. Object-based sub-pixel mapping of using local brightness and texture cues. IEEE Trans. Pattern Anal. Mach. Intell. 26,
buildings incorporating the prior shape information from remotely sensed imagery. 1–20. https://doi.org/10.1109/TPAMI.2004.1273918.
Int. J. Appl. Earth Obs. Geoinf. 18, 283–292. https://doi.org/10.1016/j.jag.2012.02. Mathieu, R., Aryal, J., Chong, A.k., 2007. Object-based classification of ikonos imagery
008. for mapping large-scale vegetation communities in urban areas. Sensors 7,
Liu, J., Li, P., Wang, X., 2015. A new segmentation method for very high resolution 2860–2880. https://doi.org/10.3390/s7112860.
imagery using spectral and morphological information. ISPRS J. Photogramm. Maxwell, T., Zhang, Y., 2006. A Fuzzy Logic Approach To Supervised Segmentation For
Remote Sens. 101, 145–162. https://doi.org/10.1016/j.isprsjprs.2014.11.009. Object Oriented Classification. In: ASPRS Annual Conference Reno. Nevada, pp. 1–5.
Liu, J., Tang, Z., Cui, Y., Wu, G., 2017. Local competition-based superpixel segmentation Mayunga, S.D., Coleman, D.J., Zhang, Y., 2007. A semi-automated approach for ex-
algorithm in remote sensing. Sensors 17, 1364. https://doi.org/10.3390/s17061364. tracting buildings from QuickBird imagery applied to informal settlement mapping.
Liu, L., 2018. A modified approach combining FNEA and watershed algorithms for seg- Int. J. Remote Sens. 28, 2343–2357. https://doi.org/10.1080/01431160600868474.
menting remotely-sensed optical images, in: SPIE 10620. In: 2017 International Meer, P., Georgescu, B., 2001. Edge detection with embedded confidence. IEEE Trans.
Conference on Optical Instruments and Technology: Optoelectronic Imaging/ Pattern Anal. Mach. Intell. 23, 1351–1365. https://doi.org/10.1109/34.977560.
Spectroscopy and Signal Processing Technology. SPIE, Beijing, China. https://doi. Mehnert, A., Jackway, P., 1997. An improved seeded region growing algorithm. Pattern
org/10.1117/12.2300543. Recognit. Lett. 18, 1065–1071. https://doi.org/10.1016/S0167-8655(97)00131-1.
Liu, L., Sclaroff, S., 2001. Region segmentation via deformable model-guided split and Meinel, G., Neubert, M., Sensing, R., City, L., 2004. A comparison of segmentation pro-
merge. In: Proceedings Eighth IEEE International Conference on Computer Vision. grams for high resolution remote sensing data. Int. Arch. Photogramm. Remote Sens.
ICCV 2001. IEEE Comput. Soc, pp. 98–104. https://doi.org/10.1109/ICCV.2001. 35, 1097–1105.
937504. Melas, D.E., Wilson, S.P., 2002. Double Markov random fields and Bayesian image seg-
Liu, W., Liang, Y., Ren, X., Duan, P., 2008. A new contour detection in mammogram using mentation. IEEE Trans. Signal Process. 50, 357–365. https://doi.org/10.1109/78.
sequential edge linking. In: 2008 Second International Symposium on Intelligent 978390.
Information Technology Application. IEEE, pp. 197–200. https://doi.org/10.1109/ Meng, Y., Lin, C., Cui, W., Yao, J., 2014. Scale selection based on Moran’s I for seg-
IITA.2008.410. mentation of high resolution remotely sensed images. IEEE Int. Geosci. Remote Sens.
Liu, Y., Bian, L., Meng, Y., Wang, H., Zhang, S., Yang, Y., Shao, X., Wang, B., 2012. Symp. 4895–4898.
Discrepancy measures for selecting optimal combination of parameter values in ob- Meyer, F., Beucher, S., 1990. Morphological segmentation. J. Vis. Commun. Image
ject-based image analysis. ISPRS J. Photogramm. Remote Sens. 68, 144–156. https:// Represent. 1, 21–46. https://doi.org/10.1016/1047-3203(90)90014-M.
doi.org/10.1016/j.isprsjprs.2012.01.007. Mezaris, V., Kompatsiaris, I., Strintzis, M.G., 2004. Still image segmentation tools for
Lizarazo, I., Elsner, P., 2011. Segmentation of remotely sensed imagery: moving from object-based multimedia applications. Int. J. Pattern Recognit. Artif. Intell. 18,
sharp objects to fuzzy regions. In: Ho, P.-G. (Ed.), Image Segmentation. InTech, pp. 701–725. https://doi.org/10.1142/S0218001404003393.
249–272. Miao, Z., Shi, W., Gamba, P., Li, Z., 2015. An object-based method for road network
Lu, D.-S., Chen, C.-C., 2008. Edge detection improvement by ant colony optimization. extraction in vhr satellite images. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 8,
Pattern Recognit. Lett. 29, 416–425. https://doi.org/10.1016/J.PATREC.2007.10. 4853–4862. https://doi.org/10.1109/JSTARS.2015.2443552.
021. Miao, Z., Shi, W., Zhang, H., Wang, X., 2013. Road centerline extraction from high-re-
Lu, Y., Jain, R.C., 1989. Behavior of edges in scale space. IEEE Trans. Pattern Anal. Mach. solution imagery based on shape features and multivariate adaptive regression
Intell. 11, 337–356. https://doi.org/10.1109/34.19032. splines. IEEE Geosci. Remote Sens. Lett. 10, 583–587. https://doi.org/10.1109/
Lucchese, L., Mitray, S.K., 2001. Color image segmentation: A state-of-the-art survey. In: LGRS.2012.2214761.
Indian National Science Academy (INSA-A). Delhi, India, pp. 207–221. https://doi. Michel, J., Grizonnet, M., Jaen, A., Harasse, S., Hermitte, L., 2012. Open tools and
org/10.1.1.84.4896. methods for large scale segmentation of very high resolution satellite images. In:
Lucieer, A., Stein, A., Fisher, P., 2005. Multivariate texture-based segmentation of re- OGRS. pp. 179–184.
motely sensed imagery for extraction of objects and their uncertainty. Int. J. Remote Michel, J., Youssefi, D., Grizonnet, M., 2015. Stable mean-shift algorithm and its appli-
Sens. 26, 2917–2936. https://doi.org/10.1080/01431160500057723. cation to the segmentation of arbitrarily large remote sensing images. IEEE Trans.
Ma, L., Cheng, L., Li, M., Liu, Y., Ma, X., 2015. Training set size, scale, and features in Geosci. Remote Sens. 53, 952–964. https://doi.org/10.1109/TGRS.2014.2330857.
Geographic Object-Based Image Analysis of very high resolution unmanned aerial Ming, D., Ci, T., Cai, H., Li, L., Qiao, C., Du, J., 2012. Semivariogram-based spatial
vehicle imagery. ISPRS J. Photogramm. Remote Sens. 102, 14–27. https://doi.org/ bandwidth selection for remote sensing image segmentation with mean-shift algo-
10.1016/j.isprsjprs.2014.12.026. rithm. IEEE Geosci. Remote Sens. Lett. 9, 813–817. https://doi.org/10.1109/LGRS.
Ma, L., Li, M., Ma, X., Cheng, L., Du, P., Liu, Y., 2017. A review of supervised object-based 2011.2182604.
land-cover image classification. ISPRS J. Photogramm. Remote Sens. 130, 277–293. Ming, D., Li, J., Wang, J., Zhang, M., 2015. Scale parameter selection by spatial statistics
https://doi.org/10.1016/j.isprsjprs.2017.06.001. for GeOBIA: Using mean-shift based multi-scale segmentation as an example. ISPRS J.
Maboudi, M., Amini, J., Hahn, M., 2016a. Objects grouping for segmentation of roads Photogramm. Remote Sens. 106, 28–41. https://doi.org/10.1016/j.isprsjprs.2015.
network in high resolution images of urban areas. Int. Arch. Photogramm. Remote 04.010.
Sens. Spat. Inf. Sci. - ISPRS Arch. 41, 897–902. https://doi.org/10.5194/ Mirghasemi, S., Rayudu, R., Zhang, M., 2013. A new image segmentation algorithm based
isprsarchives-XLI-B7-897-2016. on modified seeded region growing and particle swarp optimization. Image Vis.
Maboudi, M., Amini, J., Hahn, M., Saati, M., 2016b. Road network extraction from VHR Comput. 28, 382–387.
satellite images using context aware object feature integration and tensor voting. Mitra, P., Uma Shankar, B., Pal, S.K., 2004. Segmentation of multispectral remote sensing
Remote Sens. 8. https://doi.org/10.3390/rs8080637. images using active support vector machines. Pattern Recognit. Lett. 25, 1067–1074.
Maintz, T., 2005. Segmentation. Digital and Medical Image Processing. Universiteit https://doi.org/10.1016/J.PATREC.2004.03.004.
Utrecht. Moellering, H., Tobler, W., 1972. Geographical variances. Geogr. Anal. 4, 34–50. https://
Malenovský, Z., Bartholomeus, H.M., Acerbi-Junior, F.W., Schopfer, J.T., Painter, T.H., doi.org/10.1111/j.1538-4632.1972.tb00455.x.
Epema, G.F., Bregt, A.K., 2007. Scaling dimensions in spectroscopy of soil and ve- Mohammadzadeh, A., Zoej, M.J.V., 2010. A Self-organizing Fuzzy Segmentation (SOFS)
getation. Int. J. Appl. Earth Obs. Geoinf. 9, 137–164. https://doi.org/10.1016/J.JAG. Method for Road Detection from High Resolution Satellite Images. Photogramm. Eng.
2006.08.003. Remote Sens. 76, 27–35. https://doi.org/10.14358/PERS.76.1.27.
Mallinis, G., Koutsias, N., Tsakiri-Strati, M., Karteris, M., 2008. Object-based classification Moigne, J. Le, Tilton, J.C., Member, S., 1995. Refining image segmentation. IEEE Trans.
using Quickbird imagery for delineating forest vegetation polygons in a Geosci. Remote Sens. 33.
Mediterranean test site. ISPRS J. Photogramm. Remote Sens. 63, 237–250. https:// Mokhtarzade, M., Zoej, M.J.V., 2007. Road detection from high-resolution satellite
doi.org/10.1016/j.isprsjprs.2007.08.007. images using artificial neural networks. Int. J. Appl. Earth Obs. Geoinf. 9, 32–40.
Manjunath, B.S., Simchony, T., Chellappa, R., 1990. Stochastic and deterministic net- https://doi.org/10.1016/j.jag.2006.05.001.
works for texture segmentation. IEEE Trans. Acoust. 38, 1039–1049. https://doi.org/ Montaghi, A., Larsen, R., Greve, M.H., 2013. Accuracy assessment measures for image
10.1109/29.56064. segmentation goodness of the Land Parcel Identification System (LPIS) in Denmark.
Manousakas, I.N., Undrill, P.E., Cameron, G.G., Redpath, T.W., 1998. Split-and-merge Remote Sens. Lett. 4, 946–955. https://doi.org/10.1080/2150704X.2013.817709.
segmentation of magnetic resonance medical images: performance evaluation and Moore, A.P., Prince, S.J.D., Warrell, J., Mohammed, U., Jones, G., 2008. Superpixel lat-
extension to three dimensions. Comput. Biomed. Res. 31, 393–412. https://doi.org/ tices. In: IEEE Conference on Computer Vision and Pattern Recognition, CVPR. IEEE,
10.1006/CBMR.1998.1489. pp. 1–8.. https://doi.org/10.1109/CVPR.2008.4587471.
Marceau, D.J., 1999. The scale issue in the social and natural sciences. Can. J. Remote Moser, G., Serpico, S.B., 2008. Classification of high-resolution images based on MRF
Sens. 25, 347–356. https://doi.org/10.1080/07038992.1999.10874734. fusion and multiscale segmentation. In: International Geoscience and Remote Sensing
Mardia, K.V., Hainsworth, T.J., 1988. A spatial thresholding method for image segmen- Symposium (IGARSS). IEEE, pp. 277–280. https://doi.org/10.1109/IGARSS.2008.
tation. IEEE Trans. Pattern Anal. Mach. Intell. 10, 919–927. https://doi.org/10.1109/ 4778981.
IVS.2005.1505186. Moser, G., Serpico, S.B., Benediktsson, J.A., 2013. Land-cover mapping by markov
Marpu, P.R., Neubert, M., Herold, H., Niemeyer, I., 2010. Enhanced evaluation of image modeling of spatial-contextual information in very-high-resolution remote sensing

131
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

images. Proc. IEEE 101, 631–651. https://doi.org/10.1109/JPROC.2012.2211551. Annu. Rev. Biomed. Eng. 2, 315–337.
Mountrakis, G., Im, J., Ogole, C., 2011. Support vector machines in remote sensing: A Poggi, G., Scarpa, G., Zerubia, J.B., 2005. Supervised segmentation of remote sensing
review. ISPRS J. Photogramm. Remote Sens. 66, 247–259. https://doi.org/10.1016/j. images based on a tree-structured MRF model. IEEE Trans. Geosci. Remote Sens. 43,
isprsjprs.2010.11.001. 1901–1911. https://doi.org/10.1109/TGRS.2005.852163.
Mueller, M., Segl, K., Kaufmann, H., 2004. Edge-and region-based segmentation tech- Pong, T.-C., Shapiro, L.G., Watson, L.T., Haralick, R.M., 1984. Experiments in segmen-
nique for the extraction of large, man-made objects in high-resolution satellite ima- tation using a facet model region grower. Comput. Vision Graph. Image Process. 25,
gery. Pattern Recognit. 37, 1619–1628. https://doi.org/10.1016/j.patcog.2004.03. 1–23. https://doi.org/10.1016/0734-189X(84)90046-X.
001. Poullis, C., You, S., 2010. Delineation and geometric modeling of road networks. ISPRS J.
Munoz, X., Freixenet, J., Cufi, X., Marti, J., 2003. Strategies for image segmentation Photogramm. Remote Sens. 65, 165–181. https://doi.org/10.1016/J.ISPRSJPRS.
combining region and boundary information. Pattern Recognit. Lett. 24, 375–392. 2009.10.004.
Mylonas, S.K., Stavrakoudis, D.G., Theocharis, J.B., Mastorocostas, P.A., 2015. A region- Prewitt, J.M.S., 1970. Picture Processing and Psychopictorics. Elsevier Science, New
based GeneSIS segmentation algorithm for the classification of remotely sensed York.
images. Remote Sens. 7, 2474–2508. https://doi.org/10.3390/rs70302474. Pu, R., Landry, S., 2012. A comparative analysis of high spatial resolution IKONOS and
Najman, L., Schmitt, M., 1996. Geodesic saliency of watershed contours and hierarchical WorldView-2 imagery for mapping urban tree species. Remote Sens. Environ. 124,
segmentation. IEEE Trans. Pattern Anal. Mach. Intell. 18, 1163–1173. https://doi. 516–533. https://doi.org/10.1016/j.rse.2012.06.011.
org/10.1109/34.546254. Pu, R., Landry, S., Yu, Q., 2011. Object-based urban detailed land cover classification with
Natowicz, R., Bergen, L., Gas, B., 1995. Kohonen’s Maps for Contour and “Region-Like” high spatial resolution IKONOS imagery. Int. J. Remote Sens. 32, 3285–3308.
Segmentation of Gray Level and Color Images. In: Artificial Neural Nets and Genetic https://doi.org/10.1080/01431161003745657.
Algorithms. Springer Vienna, Vienna, pp. 360–363. https://doi.org/10.1007/978-3- Qiu, Y., Ming, D., Zhang, X., 2016. Object oriented land cover classification combining
7091-7535-4_94. scale parameter preestimation and mean-shift segmentation. In: International
Neteler, M., Beaudette, D.E., Cavallini, P., Lami, L., Cepicky, J., 2008. GRASS GIS. In: Geoscience and Remote Sensing Symposium (IGARSS). IEEE, Beijing, China, pp.
Balram, S., Dragicevic, S. (Eds.), Open Source Approaches in Spatial Data Handling. 6332–6335. https://doi.org/10.1109/IGARSS.2016.7730655.
Springer Berlin Heidelberg, Berlin, Heidelberg, pp. 171–199. https://doi.org/10. Quirita, V.A.A., Diaz, P.A., Feitosa, R.Q., Happ, P.N., Costa, G.A.O.P., Klinger, T., Heipke,
1007/978-3-540-74831-1_9. C., 2016. Metaheuristics for supervised parameter tuning of multiresolution seg-
Neubert, M., Herold, H., Meinel, G., 2008. Assessing image segmentation quality – con- mentation. IEEE Geosci. Remote Sens. Lett. 13, 1364–1368. https://doi.org/10.1109/
cepts, methods and application. In: Blaschke, T., S.L. and G.H. (Ed.), Object-Based LGRS.2016.2586499.
Image Analysis – Spatial Concepts for Knowledge-Driven Remote Sensing Radoux, J., Defourny, P., 2007. A quantitative assessment of boundaries in automated
Applications. Springer, Berlin, pp. 769–784. forest stand delineation using very high resolution imagery. Remote Sens. Environ.
Neubert, P., Protzel, P., 2012. Superpixel benchmark and comparison. In: Forum 110, 468–475. https://doi.org/10.1016/J.RSE.2007.02.031.
Bildverarbeitung. Karlsruher Instituts für Technologie (KIT) Scientific Publishing, Rashedi, E., Nezamabadi-pour, H., 2013. A stochastic gravitational approach to feature
Karlsruhe, Germany, pp. 1–12. based color image segmentation. Eng. Appl. Artif. Intell. 26, 1322–1332. https://doi.
Nichol, J., Wong, M.S., 2008. Habitat mapping in rugged terrain using multispectral org/10.1016/J.ENGAPPAI.2012.10.002.
ikonos images. Photogramm. Eng. Remote Sens. 74, 1325–1334. https://doi.org/10. Ren, X., Malik, J., 2003. Learning a classification model for segmentation. In: Ninth IEEE
14358/PERS.74.11.1325. International Conference on Computer Vision (ICCV). Marseille, France, pp. 10–17.
Nielsen, F., Nock, R., 2003. On Region Merging: The Statistical Soundness of Fast Sorting , https://doi.org/10.1109/ICCV.2003.1238308.
with Applications. In: IEEE Computer Society Conference on Computer Vision and Robinson, G.S., 1977. Edge detection by compass gradient masks. Comput. Graph. Image
Pattern Recognition. IEEE, Madison, WI, USA. https://doi.org/10.1109/CVPR.2003. Process. 6, 492–501. https://doi.org/10.1016/S0146-664X(77)80024-5.
1211447. Rosenfield, A., Davis, L., 1979. Image segmentation and image model. Proc. IEEE 67,
Nigam, I., Huang, C., Ramanan, D., 2018. Ensemble knowledge transfer for semantic 764–772.
segmentation. In: 2018 IEEE Winter Conference on Applications of Computer Vision Saba, F., Zoej, M.J.V., Mokhtarzade, M., 2016. Optimization of Multiresolution
(WACV). IEEE, Lake Tahoe, NV, USA, pp. 1499–1508. https://doi.org/10.1109/ Segmentation for Object-Oriented Road Detection from High-Resolution Images. Can.
WACV.2018.00168. J. Remote Sens. 42, 75–84. https://doi.org/10.1080/07038992.2016.1160770.
Nijland, W., Addink, E.A., De Jong, S.M., Van der Meer, F.D., 2009. Optimizing spatial Sahr, K., White, D., Kimerling, A.J., 2003. Geodesic Discrete Global Grid Systems.
image support for quantitative mapping of natural vegetation. Remote Sens. Environ. Cartogr. Geogr. Inf. Sci. 30, 121–134. https://doi.org/10.1559/
113, 771–780. https://doi.org/10.1016/J.RSE.2008.12.002. 152304003100011090.
Nikfar, M., Zoej, M.J.V., Mohammadzadeh, A., Mokhtarzade, M., Navabi, A., 2012. Sappa, A.D., 2006. Unsupervised contour closure algorithm for range image edge-based
Optimization of multiresolution segmentation by using a genetic algorithm. J. Appl. segmentation. IEEE Trans. Image Process. 15, 377–384. https://doi.org/10.1109/
Remote Sens. 6, 063592. https://doi.org/10.1117/1.JRS.6.063592. TIP.2005.860612.
Ninsawat, S., Hossain, M.D., 2016. Identifying potential area and financial prospects of Sarkar, A., Biswas, M.K., Sharma, K.M., 2000. A Simple Unsupervised MRF Model Based
rooftop solar photovoltaics (PV). Sustainability 8, 1068. https://doi.org/10.3390/ Image Segmentation Approach. IEEE Trans. Image Process. 9, 801–812.
su8101068. Schiewe, J., 2002. Segmentation of high-resolution remotely sensed data-concepts, ap-
Nock, R., Nielsen, F., 2004. Statistical region merging. IEEE Trans. Pattern Anal. Mach. plications and problems. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 34,
Intell. 26, 1452–1458. 380–385.
Ohlander, R., Price, K., Reddy, D.R., 1978. Picture segmentation using a recursive region Shackelford, A.K., Davis, C.H., 2003. A combined fuzzy pixel-based and object-based
splitting method. Comput. Graph. Image Process. 8, 313–333. https://doi.org/10. approach for classification of high-resolution multispectral data over urban areas.
1016/0146-664X(78)90060-6. IEEE Trans. Geosci. Remote Sens. 41, 2354–2363. https://doi.org/10.1109/TGRS.
Ohta, Y.-I., Kanade, T., Sakai, T., 1980. Color information for region segmentation. 2003.815972.
Comput. Graph. Image Process. 13, 222–241. https://doi.org/10.1016/0146-664X Shankar, B., 2007. Novel classification and segmentation techniques with application to
(80)90047-7. remotely sensed images. Trans. rough sets VII 295–380.
Ojala, T., Pietikäinen, M., 1999. Unsupervised texture segmentation using feature dis- Shi, C., Wang, L., 2014. Incorporating spatial information in spectral unmixing: A review.
tributions. Pattern Recognit. 32, 477–486. https://doi.org/10.1016/S0031-3203(98) Remote Sens. Environ. 149, 70–87. https://doi.org/10.1016/J.RSE.2014.03.034.
00038-7. Shi, W., Miao, Z., Debayle, J., 2014. An integrated method for urban main-road centerline
Opitz, D., Blundell, S., 2008. Object recognition and image segmentation: the Feature extraction from optical remotely sensed imagery. IEEE Trans. Geosci. Remote Sens.
Analyst® approach. In: Blaschke, T., Lang, S., Hay, G.J. (Eds.), Object-Based Image 52, 3359–3372. https://doi.org/10.1109/TGRS.2013.2272593.
Analysis. Springer, Berlin, Heidelberg, Berlin, Heidelberg, pp. 153–167. https://doi. Shih, F.Y., Cheng, S., 2005. Automatic seeded region growing for color image segmen-
org/10.1007/978-3-540-77058-9_8. tation. Image Vis. Comput. 23, 877–886. https://doi.org/10.1016/j.imavis.2005.05.
Pal, N.R., Pal, S.K., 1993. A review on image segmentation techniques. Pattern Recognit. 015.
26, 1277–1294. https://doi.org/10.1016/0031-3203(93)90135-J. Shih, F.Y., Cheng, S., 2004. Adaptive mathematical morphology for edge linking. Inf. Sci.
Pal, S.K., Pal, N.R., 1987. Segmentation based on measures of contrast, homogeneity, and (Ny) 167, 9–21. https://doi.org/10.1016/j.ins.2003.07.020.
region size. IEEE Trans. Syst. Man. Cybern. 17, 857–868. https://doi.org/10.1109/ Srivastava, M., Arora, M.K., Raman, B., 2015. Selection of critical segmentation-A pre-
TSMC.1987.6499294. requisite for Object based image classification. In: National Conference on Recent
Peña-Barragán, J.M., Ngugi, M.K., Plant, R.E., Six, J., 2011. Object-based crop identifi- Advances in Electronics & Computer Engineering. IEEE, Roorkee, India, pp. 143–148.
cation using multiple vegetation indices, textural features and crop phenology. https://doi.org/10.1109/RAECE.2015.7510243.
Remote Sens. Environ. 115, 1301–1316. https://doi.org/10.1016/j.rse.2011.01.009. Stefanski, J., MacK, B., Waske, O., 2013. Optimization of object-based image analysis
Peng, J., Zhang, D., Liu, Y., 2005. An improved snake model for building detection from with random forests for land cover mapping. IEEE J Sel. Top. Appl. Earth Obs.
urban aerial images. Pattern Recognit. Lett. 26, 587–595. https://doi.org/10.1016/j. Remote Sens. 6, 2492–2504. https://doi.org/10.1109/JSTARS.2013.2253089.
patrec.2004.09.033. Stumpf, A., Kerle, N., 2011. Object-oriented mapping of landslides using Random Forests.
Percival, D.P., 1995. On estimation of the wavelet variance. Biometrika 82, 619–631. Remote Sens. Environ. 115, 2564–2577. https://doi.org/10.1016/j.rse.2011.05.013.
https://doi.org/10.1093/biomet/82.3.619. Su, T., 2017. A novel region-merging approach guided by priority for high resolution
Perona, P., Malik, J., 1990. Scale-space and edge detection using anisotropic diffusion. image segmentation. Remote Sens. Lett. 8, 771–780. https://doi.org/10.1080/
IEEE Trans. Pattern Anal. Mach. Intell. 12, 629–639. https://doi.org/10.1109/34. 2150704X.2017.1320441.
56205. Su, T., Zhang, S., 2018. Multi-Scale Segmentation Method Based on Binary Merge Tree
Pesaresi, M., Benediktsson, J.A., 2001. A new approach for the morphological segmen- and Class Label Information. IEEE Access 6, 17801–17816. https://doi.org/10.1109/
tation of high-resolution satellite imagery. IEEE Trans. Geosci. Remote Sens. 39, ACCESS.2018.2819988.
309–320. https://doi.org/10.1109/36.905239. Su, T., Zhang, S., 2017. Local and global evaluation for remote sensing image segmen-
Pham, D.L., Xu, C., Prince, J.L., 2000. Current methods in medical image segmentation. tation. ISPRS J. Photogramm. Remote Sens. 130, 256–276. https://doi.org/10.1016/

132
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

j.isprsjprs.2017.06.003. 00015-0.
Suk, M., Chung, S.-M., 1983. A new image segmentation technique based on partition Wang, J., Aldred, D.a., 2011. A method for obtaining and applying classification para-
mode test. Pattern Recognit. 16, 469–480. https://doi.org/10.1016/0031-3203(83) meters in object-based urban rooftop extraction from VHR multispectral images. Int.
90051-1. J. Remote Sens. 32, 2811–2823. https://doi.org/10.1080/01431161003745590.
Sun, W., Messinger, D.W., 2013. Knowledge-based automated road network extraction Wang, L., Dai, Q., Hong, L., Liu, G., 2012. Adaptive regional feature extraction for very
system using multispectral images. Opt. Eng. 52, 047203. https://doi.org/10.1117/1. high spatial resolution image classification. J. Appl. Remote Sens. 6, 063506. https://
OE.52.4.047203. doi.org/10.1117/1.JRS.6.063506.
Sun, Y., He, G.J., 2008. Segmentation of high-resolution remote sensing image based on Wang, L., Dai, Q., Xu, Q., Zhang, Y., 2015a. Constructing hierarchical segmentation tree
marker-based watershed algorithm. In: 5th International Conference on Fuzzy for feature extraction and land cover classification of high resolution MS imagery.
Systems and Knowledge Discovery, FSKD 2008. IEEE, pp. 271–276. https://doi.org/ IEEE J Sel. Top. Appl. Earth Obs. Remote Sens. 8, 1946–1961. https://doi.org/10.
10.1109/FSKD.2008.249. 1109/JSTARS.2015.2428232.
Svoboda, T., Jan, K., Vaclav, H., 2007. Image Processing, Analysis & Machine Vision - A Wang, M., Cui, Q., Wang, J., Ming, D., Lv, G., 2017a. Raft cultivation area extraction from
Matlab Companion. Thomas Learning. high resolution remote sensing imagery by fusing multi-scale region-line primitive
Tarabalka, Y., Chanussot, J., Benediktsson, J.A., Angulo, J., Fauvel, M., 2008. association features. ISPRS J. Photogramm. Remote Sens. 123, 104–113. https://doi.
Segmentation and Classification of Hyperspectral Data using Watershed. In: org/10.1016/J.ISPRSJPRS.2016.10.008.
International Geoscience and Remote Sensing Symposium (IGARSS). IEEE Wang, M., Dong, Z., Cheng, Y., Li, D., 2017c. Optimal segmentation of high-resolution
International, pp. 652–655. https://doi.org/10.1109/IGARSS.2008.4779432. remote sensing image by combining superpixels with the minimum spanning tree.
Tehrany, M.S., Pradhan, B., Jebuv, M.N., 2014. A comparative assessment between object IEEE Trans. Geosci. Remote Sens. 56, 228–238. https://doi.org/10.1109/TGRS.2017.
and pixel-based classification approaches for land use/land cover mapping using 2745507.
SPOT 5 imagery. Geocarto Int. 29, 351–369. https://doi.org/10.1080/10106049. Wang, M., Huang, J., Ming, D., 2017b. Region-line association constraints for high-re-
2013.768300. solution image segmentation. IEEE J Sel. Top. Appl. Earth Obs. Remote Sens. 10,
Teodoro, A.C., Araujo, R., 2016. Comparison of performance of object-based image 628–637.
analysis techniques available in open source software (Spring and Orfeo Toolbox/ Wang, M., Li, R., 2014. Segmentation of high spatial resolution remote sensing imagery
Monteverdi) considering very high spatial resolution data. J. Appl. Remote Sens. 10, based on hard-boundary constraint and two-stage merging. IEEE Trans. Geosci.
016011. https://doi.org/10.1117/1.JRS.10.016011. Remote Sens. 52, 5712–5725. https://doi.org/10.1109/TGRS.2013.2292053.
Tilton, J.C., 2010. Split-remerge method for eliminating processing window artifacts in Wang, M., Sun, Y., Chen, G., 2015b. Refining high spatial resolution remote sensing image
recursive hierarchical segmentation. 7,697,759. segmentation for man-made objects through a collinear and ipsilateral neighborhood
Tilton, J.C., Hall, D.K., Riggs, G.A., 2010. Creation of ersatz ground reference data for model. Photogramm. Eng. Remote Sens. 81, 397–406. https://doi.org/10.14358/
validating the MODIS snow and ICE product suite. In: International Geoscience and PERS.81.5.397.
Remote Sensing Symposium (IGARSS). IEEE, Honolulu, HI, USA, pp. 2371–2374. Wang, M., Wan, Q.M., Gu, L.B., Song, T.Y., 2013. Remote-sensing image retrieval by
https://doi.org/10.1109/IGARSS.2010.5650287. combining image visual and semantic features. Int. J. Remote Sens. 34, 4200–4223.
Tilton, J.C., Tarabalka, Y., Montesano, P.M., Gofman, E., 2012. Best merge region- https://doi.org/10.1080/01431161.2013.774098.
growing segmentation with integrated nonadjacent region object aggregation. IEEE Wang, M., Wang, J., 2016. A region-line primitive association framework for object-based
Trans. Geosci. Remote Sens. 50, 4454–4467. https://doi.org/10.1109/TGRS.2012. remote sensing image analysis. Photogramm. Eng. Remote Sens. 82, 149–159.
2190079. https://doi.org/10.14358/PERS.82.2.149.
Tran, T.N., Wehrens, R., Hoekman, D.H., Buydens, L.M.C., 2005. Initialization of Markov Wang, P., Sun, G., Wang, Z., 2015. Seismic remote sensing image segmentation based on
random field clustering of large remote sensing images. IEEE Trans. Geosci. Remote spectral histogram and dynamic region merging. In: Liu, J., Sun, H. (Eds.), Ninth
Sens. 43, 1912–1919. https://doi.org/10.1109/TGRS.2005.848427. International Symposium on Multispectral Image Processing and Pattern Recognition
Tremeau, A., Borel, N., 1997. A region growing and merging algorithm to color seg- (MIPPR2015) - The International Society for Optical Engineering. SPIE. https://doi.
mentation. Pattern Recognit. 30, 1191–1203. https://doi.org/10.1016/S0031- org/10.1117/12.2209431.
3203(96)00147-1. Wang, Y., Meng, Q., Qi, Q., Yang, J., Liu, Y., 2018c. Region merging considering within-
Trivedi, M.M., Bezdek, J.C., 1986. Low-level segmentation of aerial images with fuzzy and between-segment heterogeneity: an improved hybrid remote-sensing image
clustering. IEEE Trans. Syst. Man. Cybern. 16, 589–598. https://doi.org/10.1109/ segmentation method. Remote Sens. 10, 781. https://doi.org/10.3390/rs10050781.
TSMC.1986.289264. Wang, Z., Jensen, J.R., Im, J., 2010. An automatic region-based image segmentation al-
Troya-Galvis, A., Gançarski, P., Berti-Équille, L., 2018. Remote sensing image analysis by gorithm for remote sensing applications. Environ. Model. Softw. 25, 1149–1165.
aggregation of segmentation-classification collaborative agents. Pattern Recognit. 73, https://doi.org/10.1016/j.envsoft.2010.03.019.
259–274. https://doi.org/10.1016/j.patcog.2017.08.030. Wang, Z., Lu, C., Yang, X., 2018a. Exponentially sampling scale parameters for the effi-
Troya-Galvis, A., Gançarski, P., Berti-Équille, L., 2016. Collaborative segmentation and cient segmentation of remote-sensing images. Int. J. Remote Sens. 39, 1628–1654.
classification for remote sensing image analysis. In: 23rd International Conference on https://doi.org/10.1080/01431161.2017.1410297.
Pattern Recognition (ICPR). IEEE, Cancun, Mexico, pp. 829–834. https://doi.org/10. Wang, Z., Song, C., Wu, Z., Chen, X., 2005. Improved Watershed Segmentation Algorithm
1109/ICPR.2016.7899738. for High Resolution Remote Sensing Images Using Texture. In: International
Tsai, A., Yezzi, A., Wells, W., Tempany, C., Tucker, D., Fan, A., Grimson, W.E., Willsky, A., Geoscience and Remote Sensing Symposium (IGARSS). IEEE International, pp.
2003. A shape-based approach to the segmentation of medical imagery using level 3721–3723.
sets. IEEE Trans. Med. Imaging 22, 137–154. https://doi.org/10.1109/TMI.2002. Wang, Z., Yang, X., Lu, C., Yang, F., 2018b. A scale self-adapting segmentation approach
808355. and knowledge transfer for automatically updating land use/cover change databases
Tupin, F., Roux, M., 2005. Markov random field on region adjacency graph for the fusion using high spatial resolution images. Int. J. Appl. Earth Obs. Geoinf. 69, 88–98.
of SAR and optical data in radargrammetric applications. IEEE Trans. Geosci. Remote https://doi.org/10.1016/J.JAG.2018.03.001.
Sens. 43, 1920–1928. https://doi.org/10.1109/TGRS.2005.852080. Weickert, J., 2001. Efficient image segmentation using partial differential equations and
Turker, M., Sumer, E., 2008. Building-based damage detection due to earthquake using morphology. Pattern Recognit. 34, 1813–1824. https://doi.org/10.1016/S0031-
the watershed segmentation of the post-event aerial images. Int. J. Remote Sens. 29, 3203(00)00109-6.
3073–3089. https://doi.org/10.1080/01431160701442096. Witharana, C., Civco, D.L., 2014. Optimizing multi-resolution segmentation scale using
Tzotsos, A., Argialas, D., 2006. Mseg: a Generic Region-Based Multi-Scale Image empirical methods: Exploring the sensitivity of the supervised discrepancy measure
Segmentation. Proceedings of ASPRS 2006 Annual Conference. Reno, Nevada. Euclidean distance 2 (ED2). ISPRS J. Photogramm. Remote Sens. 87, 108–121.
Tzotsos, A., Karantzalos, K., Argialas, D., 2011. Object-based image analysis through https://doi.org/10.1016/J.ISPRSJPRS.2013.11.006.
nonlinear scale-space filtering. ISPRS J. Photogramm. Remote Sens. 66, 2–16. Woodcock, C.E., Strahler, A.H., 1987. The factor of scale in remote sensing. Remote Sens.
https://doi.org/10.1016/j.isprsjprs.2010.07.001. Environ. 21, 311–332. https://doi.org/10.1016/0034-4257(87)90015-0.
Vedaldi, A., Soatto, S., 2008. Quick shift and kernel methods for mode seeking. In: Wu, H., Li, Z.L., 2009. Scale issues in remote sensing: A review on analysis, processing and
Forsyth, D., Torr, P., Zisserman A. (Eds.), European Conference on Computer Vision – modeling. Sensors 9, 1768–1793. https://doi.org/10.3390/s90301768.
ECCV 2008. Springer, Berlin, Heidelberg, Berlin, Heidelberg, pp. 705–718. https:// Wu, L., Wang, Y., Long, J., Liu, Z., 2015. A non-seed-based region growing algorithm for
doi.org/10.1007/978-3-540-88693-8_52. high resolution remote sensing image segmentation. In: YJ, Z. (Ed.), Image and
Verma, O.P., Hanmandlu, M., Susan, S., Kulkarni, M., Jain, P.K., 2011. A simple single Graphics. ICIG 2015. Lecture Notes in Computer Science. Springer, pp. 263–277.
seeded region growing algorithm for color image segmentation using adaptive Wuest, B., Zhang, Y., 2009. Region based segmentation of QuickBird multispectral ima-
thresholding. In: 2011 International Conference on Communication Systems and gery through band ratios and fuzzy comparison. ISPRS J. Photogramm. Remote Sens.
Network Technologies. IEEE, Jammu, India, pp. 500–503. https://doi.org/10.1109/ 64, 55–64.
CSNT.2011.107. Xiaohan, Y.X.Y., Yla-Jaaski, J., Huttunen, O., Vehkomaki, T., Sipila, O., Katila, T., 1992.
Villmann, T., Merényi, E., Hammer, B., 2003. Neural maps in remote sensing image Image segmentation combining region growing and edge detection. In: 11th IAPR
analysis. Neural Networks 16, 389–403. https://doi.org/10.1016/S0893-6080(03) International Conference on Pattern Recognition. Vol. III. Conference C: Image,
00021-2. Speech and Signal Analysis. IEEE. https://doi.org/10.1109/ICPR.1992.202029.
Vincent, L., Soille, P., 1991. Watersheds in digital spaces: an efficient algorithm based on Xing, J., Sieber, R., Kalacska, M., 2014. The challenges of image segmentation in big
immersion simulations. IEEE Trans. Pattern Anal. Mach. Intell. 13, 583–598. https:// remotely sensed imagery data. Ann. GIS 20, 233–244. https://doi.org/10.1080/
doi.org/10.1109/34.87344. 19475683.2014.938774.
Wang, C., Xu, W., Pei, X.F., Zhou, X.Y., 2016. An unsupervised multi-scale segmentation Yang, J., He, Y., Caspersen, J., 2017. Region merging using local spectral angle thresh-
method based on automated parameterization. Arab. J. Geosci. 9. https://doi.org/10. olds: A more accurate method for hybrid segmentation of remote sensing images.
1007/s12517-016-2683-4. Remote Sens. Environ. 190, 137–148. https://doi.org/10.1016/j.rse.2016.12.011.
Wang, D., 1997. A multiscale gradient algorithm for image segmentation using water- Yang, J., He, Y., Caspersen, J., 2016. A self-adapted threshold-based region merging
shelds. Pattern Recognit. 30, 2043–2052. https://doi.org/10.1016/S0031-3203(97) method for remote sensing image segmentation. In: International Geoscience and

133
M.D. Hossain and D. Chen ISPRS Journal of Photogrammetry and Remote Sensing 150 (2019) 115–134

Remote Sensing Symposium (IGARSS). IEEE, Beijing, China, pp. 6320–6323. Zhang, X., Xiao, P., Feng, X., 2017c. Toward combining thematic information with
Yang, J., He, Y., Caspersen, J., Jones, T., 2015a. A discrepancy measure for segmentation hierarchical multiscale segmentations using tree Markov random field model. ISPRS
evaluation from the perspective of object recognition. ISPRS J. Photogramm. Remote J. Photogramm. Remote Sens. 131, 134–146. https://doi.org/10.1016/j.isprsjprs.
Sens. 101, 186–192. https://doi.org/10.1016/j.isprsjprs.2014.12.015. 2017.08.003.
Yang, J., He, Y., Weng, Q., 2015b. An automated method to parameterize segmentation Zhang, X., Xiao, P., Feng, X., 2014a. Fast hierarchical segmentation of high-resolution
scale by enhancing intrasegment homogeneity and intersegment heterogeneity. IEEE remote sensing image with adaptive edge penalty. Photogramm. Eng. Remote Sens.
Geosci. Remote Sens. Lett. 12, 1282–1286. https://doi.org/10.1109/LGRS.2015. 80, 71–80. https://doi.org/10.14358/PERS.80.1.71.
2393255. Zhang, X., Xiao, P., Feng, X., Feng, L., Ye, N., 2015c. Toward evaluating multiscale seg-
Yang, J., Jones, T., Caspersen, J., He, Y., 2015c. Object-based canopy gap segmentation mentations of high spatial resolution remote sensing images. IEEE Trans. Geosci.
and classification: Quantifying the pros and cons of integrating optical and LiDAR Remote Sens. 53, 3694–3706. https://doi.org/10.1109/TGRS.2014.2381632.
data. Remote Sens. 7, 15917–15932. https://doi.org/10.3390/rs71215811. Zhang, X., Xiao, P., Feng, X., Wang, J., Wang, Z., 2014b. Hybrid region merging method
Yang, J., Li, P., He, Y., 2014. A multi-band approach to unsupervised scale parameter for segmentation of high-resolution remote sensing images. ISPRS J. Photogramm.
selection for multi-scale image segmentation. ISPRS J. Photogramm. Remote Sens. Remote Sens. 98, 19–28. https://doi.org/10.1016/j.isprsjprs.2014.09.011.
94, 13–24. https://doi.org/10.1016/j.isprsjprs.2014.04.008. Zhang, X., Xiao, P., Song, X., She, J., 2013. Boundary-constrained multi-scale segmen-
Yang, Y., Han, C., Han, D., 2008. A Markov random field model-based fusion approach to tation method for remote sensing images. ISPRS J. Photogramm. Remote Sens. 78,
segmentation of SAR and optical images. In: International Geoscience and Remote 15–25. https://doi.org/10.1016/j.isprsjprs.2013.01.002.
Sensing Symposium (IGARSS). IEEE, pp. 802–805. https://doi.org/10.1109/IGARSS. Zhang, Y.-J., 2006. An Overview of Image and Video Segmentation in the Last 40 Years.
2008.4779844. In: Zhang, Y.-.J. (Ed.), Advances in Image and Video Segmentation. IRM Press,
Yin, D., Du, S., Wang, S., Guo, Z., 2015. A direction-guided ant colony optimization Pennsylvania, USA, pp. 1–15.
method for extraction of urban road information from very-high-resolution images. Zhang, Y., Feng, X., Le, X., 2008a. Segmentation on Multispectral Remote Sensing Image
IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 8, 4785–4794. https://doi.org/10. Using Watershed Transformation. In: Congress on Image and Signal Processing, 2008.
1109/JSTARS.2015.2477097. IEEE, pp. 773–777. https://doi.org/10.1109/CISP.2008.365.
Yin, W., Yang, J., 2017. Sub-pixel vs. super-pixel-based greenspace mapping along the Zhang, Y., Matuszewski, B.J., Shark, L.-K., Moore, C.J., 2008b. Medical Image
urban–rural gradient using high spatial resolution Gaofen-2 satellite imagery: a case Segmentation Using New Hybrid Level-Set Method. In: 2008 Fifth International
study of Haidian District, Beijing China. Int. J. Remote Sens. 38, 6386–6406. https:// Conference BioMedical Visualization: Information Visualization in Medical and
doi.org/10.1080/01431161.2017.1354266. Biomedical Informatics. IEEE, London, UK, pp. 71–76. https://doi.org/10.1109/
Yu, Q., Clausi, D.A., 2008. IRGS: Image segmentation using edge penalties and region MediVis.2008.12.
growing. IEEE Trans. Pattern Anal. Mach. Intell. 30, 2126–2139. https://doi.org/10. Zhang, Y.J., 1997. Evaluation and comparison of different segmentation algorithms.
1109/TPAMI.2008.15. Pattern Recognit. Lett. 18, 963–974. https://doi.org/https://doi.org/10.1016/
Yu, Q., Gong, P., Clinton, N., Biging, G., Kelly, M., Schirokauer, D., 2006. Object-based S0167-8655(97)00083-4.
detailed vegetation classification with airborne high spatial resolution remote sensing Zhang, Y.J., 1996. A survey on evaluation methods for image segmentation. Pattern
imagery. Photogramm. Eng. Remote Sens. 72, 799–811. https://doi.org/10.14358/ Recognit. 29, 1335–1346. https://doi.org/10.1016/0031-3203(95)00169-7.
PERS.72.7.799. Zhao, M., Li, F., Tang, G., 2012. Optimal Scale Selection for DEM Based Slope
Yuan, J., Cheriyadat, A.M., 2013. Road Segmentation in Aerial Images by Exploiting Road Segmentation in the Loess Plateau. Int. J. Geosci. 03, 37–43. https://doi.org/10.
Vector Data. In: 2013 Fourth International Conference on Computing for Geospatial 4236/ijg.2012.31005.
Research and Application. IEEE, San Jose, CA, USA, pp. 16–23. https://doi.org/10. Zheng, C., Wang, L., Chen, R., Chen, X., 2013. Image Segmentation Using Multiregion-
1109/COMGEO.2013.4. Resolution MRF Model. IEEE Geosci. Remote Sens. Lett. 10, 816–820.
Yuan, J., Wang, D., Li, R., 2014. Remote sensing image segmentation by combining Zhengqin, Li, Jiansheng, Chen, 2015. Superpixel segmentation using Linear Spectral
spectral and texture features. IEEE Trans. Geosci. Remote Sens. 52, 16–24. https:// Clustering. 2015 IEEE Conf. Comput. Vis. Pattern Recognit. 1356–1363. https://doi.
doi.org/10.1109/TGRS.2012.2234755. org/10.1109/CVPR.2015.7298741.
Yue, A., Yang, J., Zhang, C., Su, W., Yun, W., Zhang, Y., 2011. A watershed segmentation Zhong, Y., Gao, R., Zhang, L., 2016. Multiscale and Multifeature Normalized Cut
method with shape-based merging criterion based on multispectral remote sensing Segmentation for High Spatial Resolution Remote Sensing Imagery. IEEE Trans.
imagery. Sens. Lett. 9. https://doi.org/10.1166/sl.2011.1389. Geosci. Remote Sens. 54, 6061–6075. https://doi.org/10.1109/TGRS.2016.2580643.
Zarrinpanjeh, N., Samadzadegan, F., Schenk, T., 2013. A new ant based distributed fra- Zhou, C., Wang, P., Zhang, Z., Qi, C., Wang, Y., 2007. Object-oriented information ex-
mework for urban road map updating from high resolution satellite imagery. traction technology from QuickBird pan-sharpened images. Int. Soc. Opt. Eng. 6279.
Comput. Geosci. 54, 337–350. https://doi.org/10.1016/J.CAGEO.2012.12.006. https://doi.org/10.1117/12.725360.
Zhang, A.Z., Sun, G.Y., Liu, S.H., Wang, Z.J., Wang, P., Ma, J.S., 2017a. Multi-scale Zhou, H., Kong, H., Wei, L., Creighton, D., Nahavandi, S., 2016. On Detecting Road
segmentation of very high resolution remote sensing image based on gravitational Regions in a Single UAV Image. IEEE Trans. Intell. Transp. Syst. 18, 1713–1722.
field and optimized region merging. Multimed. Tools Appl. 76, 15105–15122. https://doi.org/10.1109/TITS.2016.2622280.
https://doi.org/10.1007/s11042-017-4558-4. Zhou, Y., Feng, L., Chen, Y., Li, J., 2017a. Object-based land cover mapping using
Zhang, G., Jia, X., Hu, J., 2015d. Superpixel-based graphical model for remote sensing adaptive scale segmentation from ZY-3 satellite images. In: 2017 IEEE International
image mapping. IEEE Trans. Geosci. Remote Sens. 53, 5861–5871. https://doi.org/ Geoscience and Remote Sensing Symposium (IGARSS). IEEE, pp. 63–66. https://doi.
10.1109/TGRS.2015.2423688. org/10.1109/IGARSS.2017.8126894.
Zhang, J., Tang, Z., Gui, W., Chen, Q., Liu, J., 2017b. Interactive image segmentation with Zhou, Y., Li, J., Feng, L., Zhang, X., Hu, X., 2017b. Adaptive Scale Selection for Multiscale
a regression based ensemble learning paradigm. Front. Inf. Technol. Electron. Eng. Segmentation of Satellite Images. IEEE J Sel. Top. Appl. Earth Obs. Remote Sens. 10,
18, 1002–1020. 3641–3651. https://doi.org/10.1109/JSTARS.2017.2693993.
Zhang, L., Ji, Q., 2010. Image segmentation with a unified graphical model. IEEE Trans. Zhou, Y.T., Venkateswar, V., Chellappa, R., 1989. Edge detection and linear feature ex-
Pattern Anal. Mach. Intell. 32, 1406–1425. https://doi.org/10.1109/TPAMI.2009. traction using a 2-D random field model. IEEE Trans. Pattern Anal. Mach. Intell. 11,
145. 84–95. https://doi.org/10.1109/34.23115.
Zhang, L., Jia, K., Li, X., Yuan, Q., Zhao, X., 2014c. Multi-scale segmentation approach for Zhuowen, Tu, Zhu, Song-Chun, 2002. Image segmentation by data-driven markov chain
object-based land-cover classification using high-resolution imagery. Remote Sens. monte carlo. IEEE Trans. Pattern Anal. Mach. Intell. 24, 657–673. https://doi.org/10.
Lett. 5, 73–82. https://doi.org/10.1080/2150704X.2013.875235. 1109/34.1000239.
Zhang, L., Zhang, L., Du, B., 2016. Deep learning for remote sensing data. IEEE Geosci. Zivkovic, Z., 2004. Improved adaptive Gaussian mixture model for background subtrac-
Remote Sens. Mag. 4, 22–40. https://doi.org/10.1155/2016/7954154. tion. In: 17th International Conference on Pattern Recognition, 2004. ICPR2004. pp.
Zhang, X., Feng, X., Xiao, P., 2015a. Multi-scale segmentation of high-spatial resolution 28–31. https://doi.org/10.1109/ICPR.2004.1333992.
remote sensing images using adaptively increased scale parameter. Photogramm. Zouagui, T., Benoit-Cattin, H., Odet, C., 2004. Image segmentation functional model.
Eng. Remote Sens. 81, 461–470. https://doi.org/10.14358/PERS.81.6.461. Pattern Recognit. 37, 1785–1795. https://doi.org/10.1016/j.patcog.2003.12.014.
Zhang, X., Feng, X., Xiao, P., He, G., Zhu, L., 2015b. Segmentation quality evaluation Zuva, T., Olugbara, O.O., Ojo, S.O., Ngwira, S.M., 2011. Image Segmentation, Available
using region-based precision and recall measures for remote sensing images. ISPRS J. Techniques, Developments and Open Issues. Can. J. Image Process. Comput. Vis. 2,
Photogramm. Remote Sens. 102, 73–84. https://doi.org/10.1016/j.isprsjprs.2015. 20–29.
01.009.

134