467 
THE CONSTRUCTION OF ANTI-SYMMETRICAL WAVELET 
AND BUILDING EXTRACTION FROM REMOTE SENSING IMAGERY 
Yi LIN, Shaoming ZHANG, Feng XIE, Ying CHEN 
Department of Surveying and Geoinformatics, Tongji Univ., Shanghai 200092, China - 
linyi@mail.tongii.edu.cn 
Commission III, WG III/4, ThS-7 
KEYWORDS: Anti-symmetrical wavelet, Multi-Scale Edge Detection, Lifting wavelet, Parameter Feature Matching, 
Comer Points Recognition, 3D Building Object Extraction. 
ABSTRACT: 
The principal technical problem of 3D building semi-automatic acquisition from remote-sensing images was discussed by using 
wavelet analysis, combining digital photogrammetry and computer vision. A set of anti-symmetrical wavelet (ASWlet) functions and 
corresponding filters were deduced, e.g., multi-scales, symmetry, approximation compact support, good local regularity and 
smoothness. And it is a powerful tool to realize optimal edge detection. On the basis of the edge detection result, the improved chain 
code expression of image edge was used to describe the line feature. By using the linear template based on edge feature extraction 
and the binary image morphology we recognize the comers automatically. And then the anti-symmetric lifting wavelet algorithm to 
proceed the feature matching for getting the corresponding image points was proposed. Finally, via stereo intersection, we got the 
outline of 3D building from the stereo images. 
1. INTRODUCTION 
Extracting 3D information from objects from 2D 
remote-sensing images remains an important and essential task 
in photogrammetry vision. So, the significance of 3D 
reconstruction and expression in many projects is not only in 
theory, but also in practice. Large numbers of research on these 
aspects have been done, but the technique for fast acquiring the 
earth’s surface information and reconstructing the 3D landscape 
is still in its initial stage. The difficulty arises, of course, from 
the fact that an image is a 2D projection of the scene affected by 
noise and different defect exists in various reconstruction 
algorithms. Consequently, the knowledge of how to optimize 
and improve the automation and intelligence level of building 
reconstruction has increased the demand for photogrammetry 
vision, and has been the aim that one has been working at hard 
in the field of photogrammetry and remote sensing. 
The research on human vision mechanism and the rapid 
development of computer techniques will provide new ways for 
image feature extraction, and the research on human vision 
system shows that the spatial/frequency multi-resolution 
analysis based on wavelet transformation methods coincides 
with human vision characters. It reveals the multi-resolution, 
multi-channel characters of human vision perception processes. 
Therefore, the wavelet application in digital image processing 
has drawn intensive attention of the researchers, and today it has 
become the hotspot of the research on image feature extraction. 
In this paper, we attempt to make use of wavelet analysis, 
combining digital photogrammetry and computer vision to 
develop a critical technique of building semi-automatic 
acquisition from remote-sensing images. Therefore how to 
realize and improve the automation and intelligence level is our 
studious aim. Aimed at extracting the 3D building automatically 
or semi-automatically, we try to search for a new effective 
thought of edge feature extraction, which is according to human 
vision perception process. Here a set of Anti-symmetrical 
wavelet (ASWlet) functions and corresponding filters were 
deduced, e.g., symmetry, approximation compact support, good 
local regularity and smoothness. And it is a powerful tool to 
realize optimal edge detection. On the basis of the edge 
detection result, the improved chain code expression of image 
edge was used to describe the line feature, using the linear 
template based on wavelet edge detection to recognize the 
comers automatically. And then the anti-symmetric lifting 
wavelet was adopted to process the feature matching for 
acquiring the corresponding image points. Finally, via the stereo 
intersection we got the outline of 3D building from the stereo 
images. 
In the following sections, we will discuss the method in detail. 
2. THE CONSTRUCTION OF THE ASWLET WAVELET 
AND THE EDGE DETECTION 
2.1 Marr Hypothesis and Construction of Anti-symmetric 
Wavelet 
Marr (1982) has pointed out that describing images in human 
vision has multiresolution characteristics, whose basic meaning 
is that the description of targets provided by a retina system is 
an ordered sequence, and the scales magnitude is according to 
geometric series. Extending the hypothesis to the intensity 
detection of images, he proposed two instructive principles: 
1) the changes of an image’s intensity exist in each scale, so the 
optimized detection operator for them must be multi-scale, with 
large scales detecting the obvious outlines and small scales 
detecting the fine details of sudden changes in images; 2) 
sudden changes of intensities will result in the appearance of 
extremum in first derivative, and second derivative of a smooth 
function traverses zero point. 
Yi Lin, Ph.D, Associate professor, Department of Surveying and Geoinformatics, Tongji University.