2008
1005
A METHOD OF HIGH-RESOLUTION REMOTE SENSING IMAGES BUILDING ON
EDGE EXTRACTION
WANG DAN *, SONG WEI-dong
School of Geomatics, Liaoning Technical University, Fuxin, 123000 ,Liaoning, China —
wangdan youxiang@163.com,song wd@163.net
Commission VII, WG VII/5
KEY WORDS: Aerial Images; Building Contour; Building Recognition; Edge Extraction; Edge Extraction; Semi-Automatic
ABSTRACT:
This paper proposes a method of high-resolution remote sensing image of buildings on semi-automatic edge extraction. The first,
pre-process remote sensing image and detecting all the edge; the second, tracking the edge and extracting linear feature, extracting
the main direction line of the buildings; then using the model to judge the relations of line, segmentation and regional growth; lastly,
merge regions and extract the building contour. The paper use this method to experiment on high-resolution Quick Bird satellite
images, the result shows that this method has a higher recognition level, a better accuracy of a certain practical value.
1. INTRODUCTION
As an important element of topographic map, Buildings, whose
recognition and extraction directly affected the automatic level
of terrain feature survey. Because the building has obvious
characteristics of position, it’s recognition and precise
positioning for feature extraction, the feature matching, the
image understood, the mapping and regarded as reference body
to other targets have the vital significance. From the point of
view of practical application, the realization of remote sensing
images to identify buildings need to satisfy the remote sensing
image mapping, GIS data acquisition and automatic updates;
From the point of view of research, because of the high
diversity and complexity, the successful automatic identification
system of buildings will provide a general understanding of
guiding significance of the theory and methods. Therefore, how
to identify and extract the buildings from remote sensing images
is the one of the most important researching topics of objective
recognition.
The paper studied the computer vision, image understanding, as
well as the method of information extraction, then put forward a
semi-automatic extraction method of buildings and solved the
problem of extraction of flattened rectangular buildings.
2. METHODS OF REALIZING
Rectangular buildings or that combinations, whose obvious
characteristics in the images, is mutual orthogonal of neighbour
edges. Based on this structural characteristics of buildings,
carried out statistics to the edge of lines, obtained the main
direction of buildings, according to main line direction to get rid
of some interference, processed the main direction and its
vertical line with model, divided the image to different regional
blocks with the extension of line, chose the blocks of buildings,
merged them with adjacent regions in accordance with certain
criteria, then extracted the outside edge of combination region
and implemented linear approximation, the final outcome was
got.
2.1 Preprocessing of remote sensing image
There are a variety of influencing factors present on accessing
the remote sensing images, due to the image including
quantization noise, channel transmission noise, thermal noise
and other types of noise, and influencing the following work of
information extraction. In order to solve this problem, we need
to carry out the smoothing filter to remove the noise, however,
during the process of smoothing, the filter which requires good
ability of smoothing, at the same time , should keep the details
of images, and maintains the accuracy of marginal position. In
this paper, use adaptive smoothing method to pre-proceed the
image.
2.2 Edge detection and linear feature extraction
Building recognition and extraction on the high-resolution
remote sensing images, building top edge information is a most
important means of judge. Therefore, how to extract the top
edge of building is one of the important factors influencing the
result of extraction. In image processing, generally considered
grey changed dramatically which is the edge points. In the
frequency domain shows high-frequency component, the
process of edge detection is to locate the grey dramatic changes
in image, which is also a high-frequency enhanced process.
Compared to other classic edge detection operatorfRobert,
Sobel), Canny edge detection has good continuity and integrity,
moreover has obvious effect, after the detection, it’s necessary
to process threshold segmentation and refinement to get BW
images, easily process followed.
* Corresponding author. Wang Dan, (1983-), female, Liaoyan, Liaoning province, graduate student.
E-mail:wangdan_youxiang@163.com