Depth generator design for 3D video compression standard

• Main Authors: Min-You Hu, 何明祐
• Other Authors: Ching-Lung Su
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/67443811097950639333

碩士 === 國立雲林科技大學 === 電子工程系 === 102 === 3D high efficiency video coding (3D-HEVC) is a compression technology formulated by Joint Collaborative Team on Video Coding (JCT-VC) consisted of International Standardization Organization (ISO) and International Telecommunications Union (ITU), which is a compression standard proposed to multi-view video feature. With 3D-HEVC, multi-view video and the corresponding depth map can be compressed effectively, and depth information is applied to improve compression efficiency. 3D TV system applies encoder to compress inter frame redundancy into a bit stream. Decoder and subsequent View Synthesis Reference Software (VSRS) can let users watch 3D image on multi-view display. 3D-HEVC needs to film video sequences of different perspectives and their corresponding depth maps, which applies depth estimation algorithm to obtain depth information of images or uses cameras that can film depth information to produce their depth images. To reduce additional time for producing the corresponding depth maps or spend extra money for purchasing depth cameras, this study proposes a depth generator design for 3D video compression standard. The proposed algorithm can be mainly applied to encode which uses coding information of video sequence, such as the correlation of inter-view, to perform conditional filtering. The filtered data will be converted into depth information to achieve reduction of additional computation in encoded image and produce the corresponding depth map. Finally, to solve the blocking effect in estimated depth map, this study proposes a depth map inpainting algorithm to inpaint image with the edge line of original texture map, which makes depth map contain become more continuous in object edge. The experiment analysis shows that to compare the qualities of estimated depth maps, this study applies View Synthesis Reference Software provided by MPEG to perform image synthesis of virtual perspective with information from original image map and depth map, and compares the virtual image with the original image. Finally, in peak-signal-to-noise ratio (PSNR), the difference value between the estimated depth map and the original depth map is about -0.98 dB on average, and the coding speed decreases 2% of the entire coding time.