3-D Depth Map Estimation And Coding

The3D video technology is one of the hot topic of current research, which can reflect depth information of the scene and give users more realistic viewing experience. The Joint Video Team (JVT) proposed a3D data representation format, i.e., Multi-View-Plus-Depth (MVD). Each texture image has a depth map which is used to reflect the depth information of each point. At the receiving end, the combination of the depth map and texture image can generate a multi-view video to show the3D effect better.In fact, to obtain more accurate depth information, the process of depth map estimation should consider the object occlusion, fine structure, no matching and other factors. In addition, the quality of depth map is one of the most important factor to achieve a high-quality3D reconstruction effects. Therefore, it is necessary to ensure the decoded depth map quality when the depth map are compressed. The main work of this thesis is to study the depth map estimation and compression coding as mentioned above. The main contribution of the thesis can be concluded as following:(1) This thesis proposes a local stereo matching algorithm using rotation-skeleton-based region. The algorithm use a skeleton with45°rotation to improve the inconsistent problem of support region and texture image in traditional scheme. The experimental results show that the proposed scheme achieves more homogeneous areas.(2) This thesis proposes a depth map refinement algorithm using multi-step post-processing. The algorithm utilizes voting and interpolation, adaptive weighted median filter, hole filling method to refine the matching errors caused by object occlusion. The experimental results show that the proposed scheme reduces the error rate of depth map estimation effectively.(3) This thesis proposes a depth map coding scheme based on the correlations of intra-subband. The scheme applies brother coefficients prediction and sub-coefficients prediction to decrease coefficient value and subband entropy in high frequency subband. The experimental results show that the proposed prediction scheme keep the quality of depth map, while some bit rate can be saved.