Fast and Effective Successive Content-aware Filtering-based Upsampling for Depth Maps

• Main Authors: Yu-Ling Tseng, 曾聿齡
• Other Authors: Kuo-Liang Chung
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/5sm4m6

碩士 === 國立臺灣科技大學 === 資訊工程系 === 106 === Upsampling low resolution depth maps to high resolution maps is a necessary step before synthesizing virtual color maps for 3-D videos. In this thesis, without any training process, a fast and effective three-stage successive content-aware filtering-based (SCAF) method is proposed to upsample depth maps. The first stage of the proposed method reconstructs the missing depth pixels in the homogeneous regions by the proposed depth duplicate approach which benefits computational complexity reduction and accuracy. The second stage reconstructs the missing depth pixels in the the semi-homogeneous regions using a fast depth no-synthesis-error (D-NOSE) lookup table-based approach proposed in this study, and the reconstructed depth pixels do not cause any errors in the warped virtual color pixels. The third stage reconstructs the missing depth pixels in the non-homogeneous regions using a joint adaptive hybrid and trilateral filtering approach proposed in this study, and the reconstructed depth pixels can effectively remove edge blurring and texture copying artifacts. In addition, without any location map overhead, we propose a novel region-based reversible data hiding method. Based on sixteen typical test depth maps, thorough experiments have been carried out to demonstrate the substantial quality improvement and computational complexity reduction merits of the proposed method over the existing state-of-the-art methods.