Continuous depth map reconstruction from light fields

Light field analysis recently received growing interest, since its rich structure information benefits many computer vision tasks. In this thesis, we propose a framework to reconstruct continuous depth maps from light field data. Conventional approaches usually treat depth map reconstruction as an optimization problem with discrete labels. On the contrary, our proposed method obtains continuous depth maps without first quantizing the depth range, which preserves richer details compared with conventional discrete approaches. A structure tensor is employed to extract local depth information and its corresponding certainty levels from 2-D slices of light fields. We propose a method to refine the certainty levels of local estimations, in order to reduce the adverse effect of unreliable estimations on the optimization step. Based on the local estimations and their certainty levels, a global optimization is proposed by solving sparse linear systems. Two different affinity matrices for the linear system are employed to balance the efficiency and the quality of the optimization. Experiments on both synthetic and real light field data demonstrate the effectiveness of the proposed framework.