Stereo Video Disparity Estimation Using Multi-wavelets

Disparity estimation in stereo video processing is a crucial step in the generation of a 3D view of a scene. In this paper, a multi-wavelet based stereo correspondence matching technique for video is proposed. A multi-wavelet transform is first applied to a pair of stereo frames. Correspondence matching is initially performed at the coarsest level and relies on coarse-to-fine refinement in order to reduce the overall computational costs. Correspondence matching is carried out using a global error energy minimization technique to generate a disparity map for each of the four multiwavelet basebands of the stereo pair. Information in the resulting disparity maps is then combined using an interpolation operator to construct an initial disparity map. The information in the initial disparity map is then progressively propagated to higher resolution levels, on a coarse-to-fine basis, leading to a dense disparity map. Experimental results were generated using two sets of wide baseline convergent multi-view test videos: Breakdancers and Ballet. Results show that multi-wavelets can be a serious contender to scalar wavelets, producing smoother disparity maps with less mismatch errors compared to applying the same global error energy minimization algorithm in the wavelet domain. KeywordsMulti-wavelets, Correspondence matching, Disparity estimation, Stereo video.