Semi-global depth from focus

Reconstructing depth map from a sequence of images with different camera settings, known as Depth from Focus (DFF), has been addressed for decades. However, Classical local methods, e.g., windowed filter, usually produce uncertain estimation in textureless areas and depth discontinuities. Meanwhile, they appear sensitive to filter size and image noise. In this research a semi-global depth from focus approach that considers both the accuracy and robustness is proposed to enforce an adaptive smoothness constraint based on modified Laplacian measure. Firstly an adaptive aggregation strategy is proposed to integrate the local focus measure and then an efficient optimization method, known as scanline optimization, is introduced to approximate the global optimal of depth estimation. To evaluate the effectiveness of this research, experiments on both synthetic and real datasets are performed. The experimental results show that the proposed approach generates more robust and accurate depth maps while preserving depth discontinuities.