Quality Improvement for View Synthesis in 3D-HEVC by Post- Processing of Depth Map
Abstract— The depth map in multiview plus depth (MVD) format of 3D-HEVC is able to synthesize series virtual views in different camera positions. However, the block-based encoding structure of the video coding standard results in blocking effect especially for depth map, which not only degrades the prediction accuracy of the encoder, but also affects the video quality of the synthesized view. This paper proposes a postprocessing algorithm for depth map to preserve significant edges and smooth jagged contours resulted from blocking effect. We separate the encoded depth map into three regions by Otsu’s automatic segmentation. Then, the vision-based edge detection, which combining the JNDD (Just Noticeable Depth Difference) model, is utilized to obtain the visible depth edges in 3D display. Moreover, the geometric edges of the depth map are also acquired by the edge classification in MPEG-7. Finally, the Bilateral filter, JNDD filter and Gaussian filter are applied to the encoded depth map under various cases. The experimental results show that the proposed post-processing method obviously maintains the vital edge information of the depth map when removing the jagged contours and provides better video quality in synthesized view.
Index Terms— 3D-HEVC, depth map, post-processing, synthesized view.
Jie-Ru Lin, Mei-Juan Chen
Department of Electrical Engineering, National Dong Hwa University, TAIWAN
Cite: Jie-Ru Lin, Mei-Juan Chen, "Quality Improvement for View Synthesis in 3D-HEVC by Post- Processing of Depth Map," Proceedings of 2018 the 8th International Workshop on Computer Science and Engineering, pp. 756-760, Bangkok, 28-30 June, 2018.