Scribd
Upload
1 views

论文01 - 2001.12.Improved_techniques_for_automatic_image_segmentation

This paper presents improved techniques for automatic image segmentation using mathematical morphology, focusing on simplification, marker extraction, and boundary decision. A new marker-extraction design based on luminance and color information is proposed, along with a modified watershed algorithm to enhance boundary detection. The techniques aim to optimize image segmentation for both images and video signals.

Uploaded by

haigao
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
1 views

论文01 - 2001.12.Improved_techniques_for_automatic_image_segmentation

This paper presents improved techniques for automatic image segmentation using mathematical morphology, focusing on simplification, marker extraction, and boundary decision. A new marker-extraction design based on luminance and color information is proposed, along with a modified watershed algorithm to enhance boundary detection. The techniques aim to optimize image segmentation for both images and video signals.

Uploaded by

haigao
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 11, NO.

12, DECEMBER 2001 1273

Improved Techniques for Automatic Image Segmentation


Hai Gao, Wan-Chi Siu, and Chao-Huan Hou

Abstract—Mathematical morphology is very attractive for In this paper, we mainly deal with image segmentation, but
automatic image segmentation because it efficiently deals with the techniques are applicable to video signals.
geometrical descriptions such as size, area, shape, or connectivity
that can be considered as segmentation-oriented features. This
paper presents an image-segmentation system based on some II. SEGMENTATION STRATEGIES
well-known strategies. The segmentation process is divided into
three basic steps, namely: simplification, marker extraction, A. Watershed Algorithm and Its Extension to Deal With
and boundary decision. Simplification, which makes use of area Markers
morphology, removes unnecessary information from the image to
make it easy to segment. Marker extraction identifies the presence Several techniques have been developed for image segmen-
of homogeneous regions. A new marker-extraction design is tation. The watershed algorithm, which is an important mor-
proposed in this paper. It is based on both luminance and color phological tool for image segmentation, has been widely used
information. The goal of boundary decision is to precisely locate in recent years [3]–[5]. The watershed technique [6] is a re-
the boundary of regions detected by the marker extraction. This
gion-growing algorithm that analyzes an image as a topographic
decision is based on a region-growing algorithm which is a modi-
fied watershed algorithm. A new color distance is also defined for surface. It detects minima of the gradients of the gray-level
this algorithm. In both marker extraction and boundary decision, image and grows these minima according to the gradient values.
color measurement is used to replace grayscale measurement and It can be viewed as a flooding process. Points of contact between
L a b color space is used to replace the more straightforward the propagation originating from different minima are defined as
spaces such as the RGB color space and YUV color space.
the boundaries of the regions and are used to create the final par-
Index Terms—Image segmentation, marker extraction, mor- tition. However, the watershed algorithm often leads to extreme
phology. over segmentation [6]. There are two approaches to reduce this
problem. The first approach involves merging adjacent regions
I. INTRODUCTION according to some criteria after the use of the watershed algo-
rithm. Some recent articles such as [3]–[5] belong to this ap-

O BJECT-ORIENTED processing has recently been intro-


duced to image and video processing. MPEG-4 [1] pro-
vides a new video compression standard for multimedia ap-
proach.
The second approach extends the watershed algorithm to deal
with markers to reduce oversegmentation [7], [8]. Markers are
plications using object based coding, i.e., each video object is
a set of components marking flat regions of an image; i.e., each
coded in a separate data stream. Object-based coding provides
marker indicates the presence of an object. A marker represents
more functionality to video, like object manipulation, hyper-
the interior of an object. If the object interiors (markers) are set
links, scene composition, and the combination of natural and
to 0, and the uncertainty areas are set to 1, we get a binary marker
synthetic video. MPEG-7 [2] is a set of standardized descrip-
image. The binary marker image can be regarded as a first esti-
tors and description schemes, coding schemes for descriptions,
mation of the partition. The first estimation is not a partition it-
and a description definition language for multimedia content.
self. It contains a set of components (markers) marking the core
Examples of visual object-based descriptors are contour shape
regions, and a large number of pixels may remain unassigned.
description or object motion trajectory. Possible applications of
The next step is then to optimize the initial partition to get the
MPEG-7 are multimedia databases or authentication systems
final partition. The watershed algorithm is used for partition op-
based on automatic (unsupervised) or semiautomatic (super-
timization. We can impose the extracted markers as minima of
vised) image segmentation. Both standards use image-segmen-
the gradient function and suppress all other gradient minima [7].
tation information, but they do not describe how the segmenta-
Points of contact between the propagation originating from dif-
tion masks are derived. Thus, image or video segmentation is an
important field of research in video processing. ferent minima are defined as the boundaries of the regions and
are used to create the final partition. In this paper, we follow
the principles of the second approach and also propose a new
Manuscript received August 31, 2000; revised October 2, 2001. This paper marker-extraction design based on both luminance and color in-
was recommended by Associate Editor M. G. Strintzis.
H. Gao was with the Centre for Multimedia Signal Processing, Department formation.
of Electronic and Information Engineering, Hong Kong Polytechnic Univer-
sity, Hong Kong. He is now with the Department of Digital System Integration,
Institute of Acoustics, Chinese Academy of Sciences, Beijing, 100080, China B. Boundary Decision With a Modified Watershed Algorithm
(e-mail: [email protected]).
W.-C. Siu is with the Department of Electronic and Information Engineering,
As discussed in Section II-A, once the markers have been de-
Faculty of Engineering, Hong Kong Polytechnic University, Hung Hom, fined, the decision can be taken by the watershed algorithm. The
Kowloon, Hong Kong (e-mail: [email protected]). classical way of using the watershed algorithm to get object con-
C.-H. Hou is with the Institute of Acoustics, Chinese Academy of Sciences,
Beijing, 100080, China (e-mail: [email protected]). tours is to apply it to the morphological gradients of the signals.
Publisher Item Identifier S 1051-8215(01)11034-7. However, this method results in a great loss of information. So
1051–8215/01$10.00 © 2001 IEEE

Authorized licensed use limited to: Huawei Technologies Co Ltd. Downloaded on March 07,2023 at 09:45:47 UTC from IEEE Xplore. Restrictions apply.

You might also like