Depth Image-based Rendering with Edge-oriented Hole Filling for Multiview Synthesis and Its VLSI Implementation

• Main Authors: Tun-ChiehYang, 楊敦价
• Other Authors: Bin-da Liu
• Format: Others
• Language: en_US
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/05086794668369039244

碩士 === 國立成功大學 === 電機工程學系碩博士班 === 100 === In this thesis, an edge-oriented hole filling algorithm for depth image-based rendering is proposed by discarding the traditional smoothing depth approach. An object boundary detection method is used to detect and remove blur pixels near the object boundary of color images. With the blur pixels removal, the hole filling will become more reliable. Thus the quality of synthesis virtual view is increased. The hole after 3D warping are filled by the proposed edge-oriented hole filling algorithm for natural and smooth view syntheses. The non-smoothing depth map reduces geometric distortion while the proposed hole filling method increases hole filling accuracy. Synthesis results show the number of gates in the proposed DIBR system is 7.56k. The proposed architecture can reach 100 MHz operation frequency. The maximum frame size can achieve 720p HD (1280 × 720) or side by side FHD (1920 × 1080)@30 frames/sec. In simulation results, the proposed methods improve the synthesis virtual view quality. For performance evaluation, the PSNR can be increased from 1 dB to 4 dB while the SSIM is also close to 1.