You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Proceedings, XXth congress

Z. Y. Hang”, X. L. Chen”, Y.S. Li”, C. Q. Chen‘
? State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan
University , 129 Luoyu Road , Wuhan, 430079, CHINA, nundini@163.com
"Dept. of Civil and Structural Engineering, The Hong Kong Polytechnic University, Hong Kong, CHINA,
LED, South China Sea Institute of Oceanography, Chinese Academy of Sciences, 164 Xingangxi Road,
Guangzhou, 510301, CHINA, cqchen@scsio.ac.cn
KEY WORDS: Mathematical morphology, Multi-scale segmentation, Granulometry and anti-granulometry
This paper proposed a multi-scale segmentation method for remote sensing image based on mathematical morphology. In
mathematical morphological operations, opening transform can extract lighter connected components and closing transform can
extract darker ones with size smaller than a given structure element in a gray image. A connected component, as an object, may have a
high response to a given structure element size and a lower response to others. In this paper, granulometry and anti-granulometry were
used for detecting the most sensitive element structures of objects from a range of structure elements with different sizes.
Granulometry, an image sequence, was obtained by a series of opening transforms to the original image by using a family of structure
elements with an integral index set. Anti-granuometry was generated by closing transforms. The resulting image sequences of
granulometry and unti-granuometry were then operated by a series of derivatives , and the maximum value at each pixel corresponds
to the index of the most sensitive structure element. The index was taken as the morphological characteristic of the corresponding pixel.
The proposed segmentation method in this paper is based on the assumption that pixels with similar morphological features belong to
the same connected component. This method avoids the problems of over-segmentation and boundary pixels occurred in the classical
method of morphological segmentation.
& Lambert, 1999) .
1. INTRODUCTION Image segmentation is a key step in image processing and
analyzing. The object expression and feature detection in an
Mathematical morphology is based strictly on the mathematical image based on segmentation are required to express image in a
theory, and its initial idea is to explore the structure of image by more generalized form, which make it possible to analyze and
putting a structure element into it (Cui, 1999). The basic understand image in a higher level (Zhang, 2001). In general,
morphological operators include erosion, dilation, opening, there are two fundamentally different strategies for image
closing, which were first systematically examined by Matheron segmentation: edge detection and region growing. The standard
and Serra in 1960s, and other operators can be defined by the approach to morphological segmentation is dependent on
above basic operators. Mathematical morphology has been edge-detection (Pesaresi & Benediktsson, 1999), which
applied successfully in many fields, such as medical imaging, segments image into different regions by the edge of structure,
material sciences, and machine vision (Cui, 1999), and many but has a problem for confirming which region the boundary
attempts were related to the processing of remotely sensed pixels belong to. The approach of region growing tries to cluster
images, including segmentation (Pesaresi & Benediktsson, pixels with same or similar features into one region.
1999), feature extraction (Talbot, 1996; Vincent, 1998; Katartzis,
et.al, 2000), road network extraction on SAR image (Chanussot The most classical segmentation method of mathematical