Performance Analysis of Depth Intra Coding

Modern view synthesis methods (e.g., depth-image-based rendering or DIBR) rely on depth maps to generate additional virtual views from existing texture images. Since depth maps convey geometric scene information and are not directly observed by viewers, coding performance should be judged by the quality impact on synthesized views rather than the depth images themselves. Depth maps can be compressed using conventional intra-frame video codecs (such as H.264/AVC, H.265/HEVC, and AVS3) because they are essentially 2D images. This paper presents a comparative analysis of depth intra-coding efficiency using the reference encoders of H.264/AVC, H.265/HEVC, and AVS3. For fair evaluation, we measure each encoder’s performance based on the objective quality of a virtual view synthesized from the decoded depth data. Under consistent test conditions, experimental results show that the AVS3 and HEVC encoders achieve average bit-rate reductions of 48.06% and 43.40%, respectively, relative to H.264/AVC.