Depth Optimization for Accurate 3D Reconstruction from Light Field Images

Because the light field camera can capture both the position and direction of light simultaneously, it enables us to estimate the depth map from a single light field image and subsequently obtain the 3D point cloud structure. However, the reconstruction results based on light field depth estimation often contain holes and noisy points, which hampers the clarity of the reconstructed 3D object structure. In this paper, we propose a depth optimization algorithm to achieve a more accurate depth map. We introduce a depth confidence metric based on the photo consistency of the refocused angular sampling image. By utilizing this confidence metric, we detect the outlier points in the depth map and generate an outlier mask map. Finally, we optimize the depth map using the proposed energy function. Experimental results demonstrate the superiority of our method compared to other algorithms, particularly in addressing issues related to holes, boundaries, and noise.