Efficient Stereo Matching and Depth Enhancement Algorithms by Using Texture and Depth Consistency

摘要

The three-dimensional (3D) video is widely recognized as a visual media technique which enables viewers to perceive the depth in a scene Most 3D video standards such as 3D-HEVC all need depth information For 3D broadcasting how to accurately estimate the depth map and improve its performance the under different scene conditions becomes important In this dissertation several effective methods have been proposed to deal with the low resolution situation for depth estimation and enhancement Moreover we also discuss the different variations under the low resolution challenges such as weak texture less matching accuracy discontinuities illumination difference and occlusions and so on In other words the problems can be successfully solved without using supper-resolution techniques and multiple frames from video signals In order to realize low-resolution depth estimation and its enhancements we proposed one stereo matching algorithm and three novel depth enhancement methods to become a complete depth estimation system The complete system includes stereo matching with trinary cross color (TCC) census and triple image-based refinements advanced multilateral filters (AMF) consistency-guided filter (CF) potency guided upsampling (PGU) filter coupled with adaptive gradient fusion (AGF) filter Experimental results show that the proposed methods perform better qualities in both visual and subjective metrics than the classic methods and achieve visually comparable results to some time-consuming methods

獎項日期	2018 7月 5
原文	English
監督員	Jar-Ferr Yang (Supervisor)