Abstract:

Because the edge of the depth map can not be preserved well, and the decoded depth map impact the virtual view synthesis quality seriously, when H.264/AVC compress the depth map under uniform Q_P parameter, a new depth map coding method for virtual view synthesis based on the H.264/AVC extension JMVC (Joint Multi-view Video Coding) is introduced. According to the depth map structure characteristic, Sobel operator is used to analysis the object boundaries of the current block of depth map. If the current block contains the object boundaries, Q_P value is decreased, and if the block does not contain the object boundaries, Q_P value is increased. The lowest RDcost is chosen to be the best predict mode by computing all the 16×16,8×8 and 4×4 blocks RDcost. Experiment results indicate that using the decoded depth map to render the virtua
l view can improve the visual quality subjectively under almost the same bitrates with JMVC.

Key words: depth compression, Sobel operator, virtual view synthesis, Subjective quality