An investigation into the applicability of the wavelet transform to digital stereo image matching

Abstract

Stereo vision systems require the stereo images to be automatically matched. Multiresolution matching techniques have been applied to the matching problem. The wavelet transform, which produces a representation which is intermediate between a spatial and a frequency representation, can be viewed as a multiresolution decomposition technique. The lack of shift invariance of the wavelet transform restricts its applicability in stereo image matching. An hierarchical image matching technique is described which can use a Gaussian or a wavelet image pyramid. The results of using the two image pyramids are compared to evaluate the applicability of the wavelet transform. The results indicate that neither the wavelet pyramid or the Gaussian pyramid have an advantage over the other. However, it also indicates that the wavelet transform can be applied to digital stereo image matching.<>