423
AN EDGE DETECTION ALGORITHM FOR REMOTE SENSING IMAGE
Lixia Xue a * Zuocheng Wang b ’ c
“College of Computer Science and Technology, ChongQing University of Posts and Telecommunications,
ChongQing China-xuelx@cqupt.edu.cn;
b Software Institute, ChongQing University of Posts and Telecommunications, ChongQing; institute of Remote Sensing
and Geography Information System, Peking University, BeiJing -China-cs_wangzc@hotmail.com
Commission PS-32: WgS III/4
KEY WORDS: Multispectral RS image; Edge detection; Object-cloud; Multi-dimensional spatial mapping model
ABSTRACT:
An edge detection algorithm for multispectral remote sensing image is proposed in this paper. According to the uncertainty of the
objects in the RS image and the characters of multispectral image, we extend the one-dimensional cloud-space mapping model to the
multi-dimensional model. The object-cloud will have the multi-dimensional digital characteristics to describe the fuzziness and
randomicity of spatial objects. According to the cloud operation, multi-dimensional boundary-cloud and its digital characteristics can
be obtained and the membership matrix of transition region for each dimension can be constructed. By maximum fuzzy entropy
principle, edge detection can be accomplished in the membership matrix of transition region. Integrated the results of all dimensions
by matrix superposition, the ultimate edge map can be obtained.
1. INTRODUCTION
At present, the multispectral remote sensing image is the main
spatial data source and the integrated reflection of spectral and
geometrical character of spatial objects. It is not only the
representation of feature of chroma and brightness, but also has
complex spectral features and structure features' 1 '. To produce a
multi-dimensional image must sampling at two spatial
coordinates and the spectrum of each point in optical image, so,
the gray level of multispectral image is the function of two
spatial variables and wavelength of multi-ray 12,3 '. There is a
much difference between normal image, single spectral and
multispectral remote sensing image.
The edge detection of multispectral remote sensing image is the
important method to obtain the remote sensing information and
the base of understanding of remote sensing image. The most of
current edge detection algorithms cannot get perfect effects in
multispectral RS image. First, these algorithms are using for
normal images disposal. As for multispectral remote sensing
image, which possess the spectral features and structure features,
the theory architecture of algorithm needs to be improved.
Second, the common fuzzy edge detection algorithm base on
fuzzy sets for solving the problems of fuzziness and non-
randomicity of image. However, remote sensing image can be
seen a variable of randomicity to some extend, the accuracy of
detect result is affected 14 '; Third, remote sensing data is very
complex and mass, the efficiency of algorithm need to be
improved. The paper performs a plentiful research of the theory
architecture and algorithms of fuzzy edge detection based on
fuzzy sets theory and cloud theory. An edge detection algorithm
for multispectral RS image (MRED) is proposed based on the
detailed analysis of the characters of multispectral remote
sensing image.
2. THE MULTI-DIMENSIONAL CLOUD-SPACE
MAPPING MODEL
The multispectral remote sensing image possesses the spectral
information features and data features. There are plentiful
repetitious information and redundant data between the bands,
the relativity between the bands is not conducive to the statistic
and analysis of multispectral image' 5 ' 8 '. So, it needs to transform
the image with eigenvector firstly, mostly image information
can be centralized in few component and the relativity between
the bands can be eliminated at the same time. This method can
reduce the calculation spending, it make a sufficient preparative
for the edge detection which be performed latter' 9,10 '.
The cloud-space mapping model based on pixel level feature is
proposed in literature' 11 '. Through this model the one
dimensional or two-dimensional cloud-space can be obtained,
but the object that it acts on is not the multispectral image but
the single band image. So, if you want to execute the cloud-
based disposal to multispectral RS image, the multi-dimensional
cloud-space mapping model should be established. The two-
dimensional digital image is considered a function of gray level
of two spatial variables, whereas, the multispectral image is
generally considered as combination of a series of two-
dimensional digital images that with compact correlation. For a
RS image with m bands, the sample
setsX = |Aj, J is corresponding to a finite data set
in m -dimensional character space R" . Expression
** =/('*.■c*) expresses the multi-dimensional character
value for the point NO. k , it is a multi-dimensional vector.
Suppose there are m sub-space in universe of discourse, the
m -dimensional normal cloud can be expressed
with 3 m digital characters, just as following:
