Monoscopic Depth Map Compensation Using Recursive Method

• Main Authors: Lo, Chih-Chung, 羅至中
• Other Authors: Chang, Wen-Thong
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/90819605952046588000

碩士 === 國立交通大學 === 電信工程研究所 === 101 === This paper presents a method for depth map estimation from monoscopic video. To estimate depth from single image, some depth cues such as texture is used for initial guess. However motion vectors between current frame and previous frame is more often used. Most methods combine these cues linearly to get initial depth map estimation. However no systematic approach is available for weight estimation. Thus a recursive method is proposed for depth map estimation. After we got initial depth map, we can compute a left/right image pair by DIBR(depth image based rendering). Then this image pair can be used to compute disparity map by using DP(dynamic programming) based on stereo vision. DP takes the depth relation between pixels into account while the initial depth map estimation does not. So this disparity map get smoother depth estimation and fix the problem inside foreground objects. With this new depth map, recursive depth map compensation is used to derive a better depth estimation. Experiment results show our method can improve the original depth map and further more derive the stereo images with better visual comfort.