Light Field Depth Estimation based on Occlusion Optimization
Long Zhang, Huiping Deng, Sen Xiang, Shuang Li · 2021
Light field depth estimation has superiority for its distinctive refocus characteristics. However, it remains challenges because of presence of occlusion. In order to detect more accurate occlusion region, this paper introduces a structured forests-based edge detection method that is more consistent with human vision system, then both consistency cue in angular patch and defocus cue in refocus image are combined to estimate initial depth map, finally, a global optimization method based on Markov random field is applied to enhance the quality of the initial depth map. In the process of optimization, this paper designs an adaptive weight to protect the edge. Experiments on HCI 4D light field dataset demonstrate the proposed method can achieve sharp transition around object boundaries. The proposed method outperforms some state-of-the-art light field depth estimation methods in both qualitative and quantitative evaluations.