Research On Key Technologies Of Depth Estimation For Stereo Images

3D display can provide effective depth information,so that people can get favorable 3D visual perception.Accurate depth is the key to stereo effect for the fact that depth information can reflect the depth differences of various objects and present stereoscopic perception.The depth map obtained by the existing depth acquisition technique has the problems of low resolution and inaccurate depth value.Depth information can be extracted from image resources to obtain 3D display.Based on this background,this thesis researches deeply on two aspects,including the depth map estimation for conventional stereoscopic display and light-filed depth map estimation for true 3D volumetric display.In this paper,a novel approach based on nonlocal random walk(NRW)and motion compensation is proposed for depth estimation.NRW can restore the sharp boundary of the object while preserving the structural characteristics of the depth map.Meanwhile,motion compensation can keep temporal consistency for the moving object.The depth maps of key frames and non-key frames are obtained based on NRW and shifted bilateral filtering,respectively.In order to sharp the depth boundaries,NRW is utilized to optimize the depth sequence with texture information.Considering the temporal information of adjacent frames,motion compensation is introduced to further improve the quality of depth maps.Experimental results show that the proposed method can produce higher quality depth sequence with accurate object boundaries while maintaining consistency of sequence frames.In addition,a novel approach combining light field structure with dense stereo matching is proposed for depth estimation.Considering the method based on the light field structure can effectively keep the sharp boundary of the object,and the dense stereo matching can maintain a smooth transition in the depth region,an optimization method combining the above two approaches is proposed in the Markov random field taking occlusion into consideration.Experimental results show that the proposed method can produce high quality depth estimation results with clear depth boundary and accurate depth in background.