Enforcing spatially coherent structures in shape from focus

Most of the depth optimization schemes in shape from focus (SFF) enforce spatial coherence through convex energy functionals which oversmooth depth edges. Further, usually, no additional information about the scene is incorporated while estimating depth. In this work, we tackle the first issue by employing a nonconvex penalty that preserves depth edges effectively. For the second issue, we design a novel guidance map for SFF based on the cross-correlation between image sequence and focus volume (FV). This cross-correlation-based guidance enforces the coherent structures between the image sequence and FV. The proposed regularization framework fuses information from the guidance map, iteratively updated depth map, and the structural similarity between them. The nonconvex objective function has been solved through the majorize-minimization algorithm. An analysis has been presented that indicates the convergence behavior of the solver. Experiments have been conducted using a variety of synthetic and real image sequences. Qualitative and quantitative comparison with state-of-the-art methods indicates that the proposed method provides better depth maps.