In this paper, we propose a novel hierarchical disparity estimation/compensation (DE/DC) algorithm for stereo image coding. One way to limit the well-known drawbacks of block matching (e.g., inaccurate disparity and blocking artifacts in the decoded image) is to resort to variable size block matching (VSBM). However, VSMB may result in an inconsistent disparity estimation (i.e., such that the estimated disparity field does not correspond to the true disparity) especially as the subblock becomes small, thus leading to high frequency energy at block boundaries in the residue image. To address these problems, in this paper we propose a hybrid quadtree-based DE/DC scheme, where a Markov Random Field (MRF) model is used in combination with VSBM and selective overlapped disparity compensation to improve the disparity field consistency and lead to higher coding efficiency. Our experimental results demonstrate that the proposed block segmentation scheme achieves a higher PSNR and a more consistent disparity field, as compared to conventional VSBM schemes.
展开▼