Depth from Focus Enhancement Through Joint Filtering

In this paper, we propose to apply weighted least squares (WLS) for depth from focus enhancement. The proposed method attempts to validate those depth estimates that comply with the geometric information of the object. By taking the all-in-focus image as the guidance image in joint filtering, we transfer the structural details of the object contained in this image to the depth map. This structural information enforces the regularization of depth map in a spatially varying manner. Thus, by employing joint filtering through WLS regularization, we recover better depth maps. The experimental results for synthetic and real image sequences show the effectiveness of the proposed scheme.