International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169

Volume: 6 Issue: 6 206 - 212

______________________________________________________________________________________
A Content Based Region Separation and Analysis Approach for SAR Image
Classification

Hari Singh Barath1, Er. Jasdeep Singh Mann2

P.G. Student, Department of Computer Engineering, BMS Engineering College,
Sri Muktsar Sahib, India1
Associate Professor, Department of Computer Engineering, BMS Engineering College,
Sri Muktsar Sahib, India2

Abstract:- SAR images are the images captured through satellite or radar to monitor the specific geographical area or to extract any information regarding the
geographical structure. This information can be used to recognize the land areas or regions with specific features such as identification of water area or flood area
etc. But the images captured from satellite covers larger land regions with multiple scene pictures. To recognize the specific land area, it is required to process all
the images with defined constraints to identify the particular region. The images or the image features can be trained under some classification method to
categorize the land regions. There are various supervised and unsupervised classification methods to classify the SAR images. But the SAR images are high
resolution images with multiple region types in same images. Because of this, the existing methods are not fully capable to classify the regions accurately. There is
the requirement of more effective classification that can identify the land regions more adaptively.
Keyword: SAR
__________________________________________________*****_________________________________________________

1. INTRODUCTION environment. The rain, fog and other climate imbalance

SAR images are the satellite captured images situations includes the different kind of distortion in these
images. These all affects the image capturing and includes
which are having various associated applications to perform
the noise within the image. The problem in the capturing can
the region classification or the identification of some event
be in terms of noise, contrast unbalancing, color
or activity on some part of earth. As the images are captured
from distances and through satellite, because of this these unbalancing etc. The quality of the camera device also
images are of larger size with high resolution. An image is affects the image quality. A low quality camera can result a
low quality images. The capturing in night and in extra
able to cover the Kilometers of regions and because of this
lighting also disrupts the image images and the quality of
more effective, sensitive and accurate processing is required
the image capturing.
to process on SAR images. SAR images are having number
of associated challenges and the application areas where the Another associated problem with SAR images is
SAR image processing can be applied. In this chapter all the the larger covered space captured by the SAR images.
associated application, challenges and criticalities are Sometimes the captured region is in few meters to several
discussed. This presented work is focused on the kilometers. Because of this it becomes difficult to map the
classification of SAR images in some of the defined classes. image characterization to the actual image. If the quality of
The paper has defined the experimentation on SAR images image is poor then it becomes more difficult to process this
on images captured from the secondary sources. The larger space image. When the image is captured, sometimes
complete work is here divided in separate chapters. This a captured image can be multiple mixed regions in it. The
chapter is defined to explore the SAR image classification same image can have the water region, land region, hills etc.
and associated features. Because of this, it becomes difficult to measure the region.
The quantification is also required to identify the size of
SAR images are basically defined to classify the
each particular region so that the relative processing can be
particular terrain based on the featured analysis. As the
images are captured from distance and with high resolution, applied over the image. The region segmentation is required
to process this mixed region images.
It becomes difficult to process the SAR images. The
classification of the SAR images can be done at the internal As the SAR images represents the land images and
level i.e. to perform the individual region identification there are different forms of land cover images including the
within the same image as well as the external image rivers, mountains, ponds, building, forests etc. There are
processing to divide the available images in the separate 100s of the region variety that also different according the
image classes. population density, country and the city. Because of this,
As the SAR images are captured from distance there is the requirement of some intelligent mapping map
that can identify the region more effectively and accurately.
satellite because of this, the capturing of these images are
SAR image processing is very much dependent on the
not much pure. The purity of these images depends on the
associated application for which the processing is defined.
206
IJRITCC | June 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 6 206 - 212
______________________________________________________________________________________
Such as the identification of fire in forest is completely divided in three main stages. The work model applied in this
different to search for the water region. The separate set of work is shown here in figure-
features and different processing map is required to work on
different applications and different image classes.
Stage 1 : Region Segmentation
2. METHODOLOGY
s Clustering
In this work, an improved distance weight adaptive
model is presented to analyze the regions effectively. The
work is divided in three main stages.
In the first stage of this model, the moment analysis is
performed under the intensity value observation to identify Stage 2 : Region Bound Analysis
the area difference. Curvic Analysis
Along with this, the clustering model is applied to
separate the region. Once the region is identified, the second
stage of this work is to identify the region boundaries. The
curvic analysis is here applied to generate the regions.
Once the clear regions are obtained, the parameter
specific observations is applied on Stage 3 : Classification
different regions so that the class of the particular SAR Each region is analyzed
image will be obtained.
comparat -ively to identify the
region class

Proposed Model

2.1 IMAGE SEGMENTATION

Segmentation is a fundamental task in image
processing used to extract the region from images based on
application requirement or the process requirement. In case
of SAR images, the geographical regions can be extracted
effectively using segmentation. The segmentation is
required to identify the regions with particular
characteristics, change identification, geographical damage
identification etc. The characterization to the region can be
done based on the pixel specification, featured region
extraction and the extracted region block identification. The
segmentation can be widely divided in two sub categories
listed below.
2.1.1 REGION BASED METHOD
Region based methods basically adopt the
correlation technique and are easy to calculate and
implement but required large amount of calculations which
is a serious problem.
2.1.2 FEATURES BASED METHOD
Different feature forms are available to describe the
Flow Chart for Proposed Work image characteristics in more effective form. The region
bound identification, corners identification, statistical
In this work, an improved distance weight adaptive model is observations, colour and shape constraint are some common
presented to analyze the regions effectively. The work is form of feature representation.

207
IJRITCC | June 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 6 206 - 212
______________________________________________________________________________________
2.2CLUSTERING sliding window based intensity improvement method. The
The clustering methods are basically defined to divide algorithm for this stage is given here under-
the available data in smaller sets. Each set where signify the
Fileration(SARImg)
specific characterization. The clustering here identified as
the distance based comparison so that the clusters can be /* SARImg is the considered input image which is been
formed. The adapted process for segment member analyzed for feature extraction*/
identification is given here under {
 Generate the number of segments or partitions with
Stage 1.1 Define the Sliding window size and the
specification of centralized member
content to improve the feature.
 Process each element of data and identify the
distance from the centralized member of each
partition. Step 1. [Read the SAR Image Horizontally]
 Identify the partition which is most closed to the For i=1 to SARImg.Width
particular data element {
 Now take the average distance from all centralized Step 2. [Read the SAR Image Vertically]
elements For j=1 to SARImg.Height
Switch the cluster centre and repeat the process till {
the optimized clustering not achieved. Step 3. [Slide the winodw over the image and
extract relative winodw area]
2.3 CLASSIFICATION ExtractedBlock=Extract(SARImg,WindowSize)
The main utility of SAR image processing system is Step 4. [Apply the sliding winodw matrix over the
to identify the image class based on the region processing in image block to improve the feature]
featured mode. The relative decision is required to process ExtractedBlock = ExtractedBlock x Window
on SAR image to identify the image class. Once the Step 5. [Enhance the Block Features so that the enhanced
segmentation is applied over the image and the image featured over the block will be obtained]
features are obtained, the next work is to apply the FeaturedBlock =ImproveFeatures(ExtractedBlock)
classification algorithm. The classifier is defined under the Step 6. [Include the Block in new improved Image]
feature processing to divide the image in various region ReformedImage.Add(FeaturedBlock)
classes. These classes are based on the type of land cover. }
Some of the common classes for SAR image processing Return ReformedImage
include }
 Water Area with specification of Sub Areas Stage 2: Feature Extraction

o Rivers This stage is basically defined to extract the image features

based on which the classification can be done. The
o Ponds extraction of features is here based on the window adaptive
o Floods analysis applied over the featured image. The algorithmic
 Tree Regions with specification of Sub Areas model for feature extraction for SAR image is given here
o Forests under-
o Mountains ExtractFeatures(SARImg)
 Buildings /* In this section, the statistical features over the SAR image
o Office Areas is extracted based on which recognition is performed */
o Residential Societies {
2.4 ALGORITHM
Stage 2.1 Define the size of window respective to which
In this work two stages for Algorithm approaches the extraction will be
are suggested. formed.
Stage 1: Fileration Step 1. [Obtain the SAR Image extraction with
specification]
The first stage of this work is to apply the For i=1 to SARImg.Width
improvement to the input image based on the adaptive {
features. The feature exploration is here applied using Step 2. [Read the image]
208
IJRITCC | June 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 6 206 - 212
______________________________________________________________________________________
For j=1 to SARImg.Height
{
Step 3. [Obtain the window area over the SAR image]
SARImg=GetWindow(SARImg,Window)
Step 4. [Generate the featue over the image based on the
convolutional winodw]
FeatureImage=ApplyWindow(SARImg,Window)
Step 5. [Extract the feature image from the image]
Features=ExtractFeatuers(FeatureImage)
Step 6. [Obtain the statistical featurs from the images]
Featureset.Add(Features.Entropy)
Featureset.Add(Features.Frequency)
Featureset.Add(Features.Distance)
}
} Sample Image I
Return Featureset
} Here figure is showing the results obtained for the
3. RESULTS AND DISCUSSION sample image. The figure is showing a raster image taken
3.1 DATA COLLECTION:- from the web. The input image can be any of the satellite
image taken from the web source. The figure is showing the
Data collection is the primary requirement of such qualified image under the SAR image specifications. The
classification model on which the implementation can be image is color with multiple regions. It is required to
applied. The presented work is defined to classify the identify the most qualified region over the image.
regions over the SAR images, for this there is the
requirement to collect the radar images. For any image
processing application different primary and secondary
sources are available to collect the data.

In this present work the required data is collected from

the secondary sources In this work, the high resolution
building images are extracted from web source. The
properties of collected dataset are given here under

Properties Values

Image Type Satellite Images

Histogram Analysis (Sample 1)
Type Color
Here figure is showing the histogram analysis
Size 600x600 applied over the image. The figure is showing the intensity
level observations applied over the image. The figure is
Format JPG showing the frequency variation and the pixel strength
analysis applied over the image. Here x axis is showing the
Number of images 50 intensity range and y axis is showing the frequency of each
intensity value applied over the image.
DATA SET PROPERTIES TABLE
3.2.SIMULATION RESULTS : SAMPLE IMAGE I

209
IJRITCC | June 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 6 206 - 212
______________________________________________________________________________________
Here figure is showing the identified regions in
color form. The figure is showing the generated color map
along with region level segmentation so that the region
segmentation and the quantification will be done more
accurately.

3.3 ANALYTICAL RESULTS

Here figure is showing the pixel level analysis

applied over the image relative to three different regions.
The figure is showing the contribution of each segmented
region over the image. The highest region area is here
Iterative Window Based Segmentation (Sample 1) considered as the classifier map so that the relatively higher
effective region class is identified.
Here figure is showing the Iterative process applied
to perform window based segmentation over the image. This
window segmentation is generated so that the relative region 200000
identification will be done more effectively. The window
segmentation is analyzed under sliding mode. 150000
Water Area
100000
Land Area
50000 Mountain Area

0
1.jpg 3.jpg 5.jpg 7.jpg 9.jpg

Region Classification
3.4 COMPARATVIE ANALYSIS
 In the existing work a moment based analysis is
Region Class Identification (Sample I) performed over the image. According to this
approach a 3x3 size window is moved over the
Here figure is showing the region class based image to identify the pixel based analysis.
identification so that the region segmentation is done. The
figure is showing the region level segmentation in the form  The work cannot be completed in single visit of
of gray images. sliding window, as the analysis and colorization
process over the image.
In the existing work, the comparative analysis is been
performed under the histogram analysis. But in this work,
we have used some analytical parameters such as entropy,
visibility, standard deviation etc.
In this table the comparative accuracy analysis of proposed
work is provided against each type of SAR image and the
average accuracy. The table shows that the proposed method
has improved the accuracy against each type of SAR image.

Colored Region Identification (Sample I)

210
IJRITCC | June 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 6 206 - 212
______________________________________________________________________________________
Existing REFERENCES
Area type [1] A. S. Ismail, X. Gao and C. Deng, "SAR image
GLCM Gabor LPD Propo
Based classification based on texture feature fusion," Signal and
sed
Metho Information Processing (ChinaSIP), 2014 IEEE China
Summit & International Conference on, Xi'an, 2014, pp.
d
153-156.
Water 36.09 60.27 80.41 83.33 [2] A. T. S. Ho, E. S. Seumahu, Siu Chung Tam, Teck Wei
Chin and Hock Seng Lim, "Improving SAR image
Land 69.71 84.77 88.73 90.91
classification in tropical region through fusion with SPOT
Building/ 45.81 81.22 79.18 80 data," Geoscience and Remote Sensing Symposium
Mountains Proceedings, 1998. IGARSS '98. 1998 IEEE International,
Seattle, WA, 1998, pp. 1596-1598 vol.3.
Average 50.54 75.42 82.77 84.75
[3] B. Liu, H. Hu, K. Wang, X. Liu and W. Yu, "A number-of-
Accuracy
classes-adaptive unsupervised classification framework for
SAR images," Geoscience and Remote Sensing Symposium
Comparative Analysis (IGARSS), 2011 IEEE International, Vancouver, BC, 2011,
pp. 3799-3802
The evaluation is provided against the GLCM, Gabor and
[4] C. He, T. Zhuo, D. Ou, M. Liu and M. Liao, "Nonlinear
LPD based method. Each of the class of SAR image and the
Compressed Sensing-Based LDA Topic Model for
average accuracy evaluation is shown in the graph. The Polarimetric SAR Image Classification," in IEEE Journal of
comparative results show that method improved the overall Selected Topics in Applied Earth Observations and Remote
accuracy. Sensing, vol. 7, no. 3, pp. 972-982, March 2014.
[5] D. Antonypandiarajan and S. N. Nisharani, "Accuracy
100 assessment for SAR image classification using laws mask
80 features performance analysis of SAR image
GLCM classification," Emerging Trends in Science, Engineering
60
and Technology (INCOSET), 2012 International
40 Conference on, Tiruchirappalli, Tamilnadu, India, 2012, pp.
20 Gabor 150-153
0 [6] D. Samanta and G. Sanyal, "A Novel Approach of SAR
Image Classification Using Color Space Clustering and
LPD Based Watersheds," Computational Intelligence and
Method Communication Networks (CICN), 2012 Fourth
International Conference on, Mathura, 2012, pp. 237-240
Proposed [7] D. Xiang, T. Tang, L. Zhao and Y. Su, "Superpixel
Generating Algorithm Based on Pixel Intensity and
Location Similarity for SAR Image Classification," in IEEE
Geoscience and Remote Sensing Letters, vol. 10, no. 6, pp.
1414-1418, Nov. 2013
Comparative Analysis [8] F. Liu, L. Jiao, B. Hou and S. Yang, "POL-SAR Image
Classification Based on Wishart DBN and Local Spatial
4. CONCLUSION
Information," in IEEE Transactions on Geoscience and
SAR images are the taken from satellite or considered Remote Sensing, vol. 54, no. 6, pp. 3292-3308, June 2016
as the radar images to obtain the region level observation. [9] Fulong Chen, Chao Wang, Hong Zhang, Bo Zhang and Fan
This observation is effective to identify the application Wu, "SAR images classification using case-based reasoning
specific observations such as identification or fire regions method," 2007 IEEE International Geoscience and Remote
Sensing Symposium, Barcelona, 2007, pp. 2048-2051.
etc. The work is defined in three main stages. At the earlier
[10] G. Dong-dong, T. Tang, Y. Li and J. Lu, "Local pattern
stage, the region level map is applied to separate the
descriptor for SAR image classification," Synthetic
individual regions. Once the regions are quantified, the next Aperture Radar (APSAR), 2015 IEEE 5th Asia-Pacific
work is to generate the region curves for clear separation of Conference on, Singapore, 2015, pp. 764-767
regions. Finally, the featured comparison is applied to [11] J. Feng, L. C. Jiao, X. Zhang and D. Yang, "Bag-of-Visual-
generate the region class. The work is applied on random Words Based on Clonal Selection Algorithm for SAR
sample set. The observations show that the model has Image Classification," in IEEE Geoscience and Remote
provided the accurate region identification. Sensing Letters, vol. 8, no. 4, pp. 691-695, July 2011.
[12] J. Feng, Z. Cao and Y. Pi, "Amplitude and texture feature
based SAR image classification with a two-stage

211
IJRITCC | June 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 6 206 - 212
______________________________________________________________________________________
approach," 2014 IEEE Radar Conference, Cincinnati, OH, Transactions on Geoscience and Remote Sensing, vol. 52,
2014, pp. 0360-0364. no. 4, pp. 2197-2216, April 2014
[13] J. Xu, Z. Li, B. Tian and Q. Chen, "Improved subspace [25] T. Wang, Y. Li and H. Xiong, "A novel locally active
method for fully polarimetric SAR image classification," learning method for SAR image classification," 2014 IEEE
2013 IEEE International Geoscience and Remote Sensing Geoscience and Remote Sensing Symposium, Quebec City,
Symposium - IGARSS, Melbourne, VIC, 2013, pp. 2454- QC, 2014, pp. 4596-4599
2456.
[14] K. Kayabol and B. Gunsel, "SAR image classification with
normalized gamma process mixtures," 2013 IEEE
International Conference on Image Processing, Melbourne,
VIC, 2013, pp. 320-324.
[15] K. S. Chen, Y. C. Tzeng, C. T. Chen and J. S. Lee,
"Filtering effects on polarimetric SAR image
classification," Geoscience and Remote Sensing, 1997.
IGARSS '97. Remote Sensing - A Scientific Vision for
Sustainable Development., 1997 IEEE International, 1997,
pp. 1199-1201 vol.3
[16] L. Chen, W. Yang, Y. Liu and H. Sun, "Feature evaluation
and selection for polarimetric SAR image classification,"
IEEE 10th INTERNATIONAL CONFERENCE ON
SIGNAL PROCESSING PROCEEDINGS, Beijing, 2010,
pp. 2202-2205
[17] L. Gao, F. Pan and X. Li, "A New Fuzzy Unsupervised
Classification Method for SAR Images," 2006 International
Conference on Computational Intelligence and Security,
Guangzhou, 2006, pp. 1706-1709.
[18] L. Zhang, L. Sun, B. Zou and W. M. Moon, "Fully
Polarimetric SAR Image Classification via Sparse
Representation and Polarimetric Features," in IEEE Journal
of Selected Topics in Applied Earth Observations and
Remote Sensing, vol. 8, no. 8, pp. 3923-3932, Aug. 2015
[19] L. Zhang, X. Wang, L. Sun and W. M. Moon, "Contextual
sparse representation and its application in polarimetric
SAR image classification," IET International Radar
Conference 2015, Hangzhou, 2015, pp. 1-5.
[20] N. Talla Tankam, J. Fotsing, A. Dipanda and E. Tonye,
"SAR Image Classification Combining Structural and
Statistical Methods," Signal-Image Technology and
Internet-Based Systems (SITIS), 2011 Seventh
International Conference on, Dijon, 2011, pp. 468-475.
[21] R. Devi and A. Kumar, "A region based SAR images
classification using Improved Moment Analysis (IMA),"
Optimization, Reliabilty, and Information Technology
(ICROIT), 2014 International Conference on, Faridabad,
2014, pp. 341-346.
[22] S. Gou, X. Qiao, X. Zhang, W. Wang and F. Du,
"Eigenvalue Analysis-Based Approach for POL-SAR
Image Classification," in IEEE Transactions on Geoscience
and Remote Sensing, vol. 52, no. 2, pp. 805-818, Feb.
2014.
[23] S. P. Gou, W. F. Wang, L. C. Jiao, S. Wang and X. R.
Zhang, "Pol-SAR image classification using eigenvalue-
based joint statistical framework," 2014 IEEE Geoscience
and Remote Sensing Symposium, Quebec City, QC, 2014,
pp. 2810-2813.
[24] S. Uhlmann and S. Kiranyaz, "Integrating Color Features in
Polarimetric SAR Image Classification," in IEEE

212
IJRITCC | June 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________