Anti-occlusion And Anti-noise Depth Estimation For Light Field

As a new multi-view photography equipment,a light field camera can record space and angular information of a scene within one shot.Multi-view and refocusing images can be obtained from light field cameras,which possess unique advantage especially in depth estimation.Occlusion is a challenging issue for light field depth estimation.Previous works have failed to model occlusion or have considered only single occlusion,thereby failing to achieve accurate depth for multi-occlusion.Due to the influences of light,texture,and resolution,real light field images captured by camera may contain some noises,which greatly reduce depth estimation accuracy.Previous works have often focused to synthetic dataset without noises,but ignored the research on real noisy dataset.In this thesis,we focus on occlusion and noises which are challenging in light field depth estimation.The main work of this thesis is listed as follows:(1)We summarize the background and research status of light field camera and depth estimation.Furthermore,we introduce the basic theory of light field and the extraction method of depth through light field.(2)To solve the problem about occlusion,we define occlusions into non-occlusion,single-occlusion,and multi-occlusion types.Then we build adaptive correspondence cost volume,and present a light field depth estimation algorithm that is robust to different occlusion types.Experimental results show that the proposed approach can effectively solve the occlusion problem and obtain highly accurate edge-preserved depth map.In addition,the running-time efficiency outperforms that of other methods.(3)For noises in depth estimation,the theory of information entropy is adopted to build adaptive defocus cost volume.Furthermore,we combine correspondence and defocus cost volume weighted by their respective confidence measures.We present a robust light field depth estimation algorithm for noisy scene with occlusion.Compared with other state-of-the-art methods,the proposed method is more robust and achieves high quality depth maps in various scenes.