Multiresolutional representation such as pyramidal structures is usefu
l for stereo matching as coarse-to-fine strategy. However, conventiona
l pyramidal structures using Gaussian or Laplacian filters lose much i
nformation due to their low-pass filtering characteristics and also ca
nnot obtain any spatial orientation selectivity. The adoption of wavel
et transform can remedy these problems, but at the time of image trans
lation, it changes wavelet coefficients. In this paper, a pyramid usin
g modified wavelet decomposition process is proposed to have translati
on invariance. The image transformed by the proposed method is convert
ed into appropriate multiple features without loss of information. Sin
ce the importance of each feature is determined heuristically in the m
ultiple feature-based stereo matching method, it is very difficult to
fuse them adequately. In the proposed algorithm, the weight of each fe
ature, that is, the relative importance of each feature, is decided fr
om the similarity between the intensities in the local region of each
left and right wavelet channels. Since the window size used for the de
cision of weight and disparity values greatly influences the processed
result, the window is adaptively determined from the disparities esti
mated in the coarse resolution and low-varying channel of fine resolut
ion. The window size must be large enough to obtain signal-to-noise ra
tio, but not too large as to induce the effects of projective distorti
on. Also, a new relaxation algorithm which can reduce false matches wi
thout blurring the disparity edge is proposed. By integrating adaptive
weight variable window selection method, and relaxation process, an a
ccurate and stable disparity map is obtained. Experimental results for
various images show that the proposed algorithm has good performance
even if the image has the unfavorable conditions. (C) 1997 Pattern Rec
ognition Society.