Improving Depth Compression in HEVC by Pre/Post Processing

Depth images have different characteristics from that of color images. They usually have gradual changes within objects while steep changes happen around object boundaries. Compression standards such as H.264/AVC and High Efficiency Video Coding (HEVC) are efficient in dealing with the gradual change regions but usually result in poor performance at edge regions. To facilitate the reuse of the current video coding design and to further improve the depth compression performance, we propose a pre/post processing based compression strategy. By modifying the edge blocks in the depth image to flat blocks, the pre-processed image can be efficiently compressed using existing compression schemes. Meanwhile, those edge blocks are compressed by an edge preserving codec separately. At the decoder, the decoded modified image and the edge blocks will be merged together to form the final reconstructed image. In our simulations, we implement this strategy to HEVC to evaluate the coding performance. Experimental results show the proposed scheme can achieve about 30% - 40% bit savings for Ballet and Break dancers sequences and 60% - 70% bit savings for Kinect captured depth sequences in comparison with HEVC.