| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 104 === With the advance of the multimedia codec technologies and related application devices, the cost to transmit media data can be reduced while the system can still maintain high quality reconstructed video. In additional to the conventional 2-D multimedia applications, the Free-View Video (FVV) and 3D Video become popular in that better user perception experiences can be provided. The FVV and 3D-Video related applications have been wildly developed and utilized to drive more user interests. The FVV and 3-D Video are acquired by utilizing multi cameras from different view angle of the same scene. To reduce the costs of multi-view data transmission and enhance the quality of the reconstructed virtual views, not only the texture but also the depth information have to be well manipulated, as the latter plays an important role in virtual view synthesis. In this thesis, we propose a new depth map coding scheme to improve the inefficiency of applying conventional texture information based video coding tools on video depth map. Based on the fact that the depth map comprises sharp edges that surround object region boundaries and all depth map values belonging to the same region are nearly constant, the conventional 2D video codec which are designed to well compress video texture information is proved to be inefficient to encode depth map information. By utilizing the high correlation between texture and depth map, we propose a texture segmented region-based depth coding scheme to reserve accurate depth information while saving required bit rates. We proposed to utilize a two-layer depth image coding method, together with chain code and region segmentation results, to reduce the total FVV coding complexity. Experimental results show that the proposed method can largely shorten the encoding time and the decoded depth information can be used to improve the texture image coding performance. It helps to improve the overall FVV encoding performance.
|
|---|
