SSD Matching Using Shift-Invariant Wavelet Transform

The conventional area-based stereo matching algorithm suffers from two problems, the windowing problem and computational cost. Multiple scale analysis has long been adopted in vision research. Investigation of the wavelet transform suggests that -- dilated wavelet basis functions provide changeable window areas associated with the signal frequency components and hierarchically represent signals with multiresolution structure. This paper discusses the advantages of applying wavelet transforms to stereo matching and the weakness of Mallat’s multiresolution analysis. The shift-invariant dyadic wavelet transform is exploited to compute an image disparity map. Experimental results with synthesised and real images are presented.