Stereo data compression is an important issue for the new generation of vision systems. In this paper, we are interested in lossless coding methods for stereo images allowing progressive reconstruction. Most of the existing approaches account for the mutual similarities between the left and the right images. More precisely, the disparity compensation process consists in predicting the right image from the left one based on the disparity map. Then, the disparity map, the reference image, and the residual image are encoded. In this work, we propose a novel approach based on the concept of vector lifting scheme. Its main feature is that it does not generate one residual image but two compact multiresolution representations of the left and the right views, driven by the underlying disparity map. Experimental results show a significant improvement using this technique compared with conventional methods.
展开▼
