Abstract
This paper presents a hierarchical stereo matching strategy using the Discrete Wavelet Transform (DWT). Both area- and feature-based methods are combined into a single process by taking advantage of the discrete wavelet decomposition. Image components extracted from the approximation, horizontal, and vertical channels of the decomposition are combined to do the matching at each level. The disparity at a coarse level is then propagated to the finer levels. Because the detail channel of the decomposition is discarded, the noise is automatically reduced. Experiments using various kinds of image pairs show that the method is accurate, fast, and highly robust to noise.