Incorporating structural prior for depth regularization in shape from focus

Abstract

Depth map reconstructed through passive or active methods usually has noise and exhibits poor shape quality. In the case of shape from focus (SFF), improvement techniques can be divided mainly into two categories: one, which enhances the image focus volume, and second, which tries to refine the depth map. In both of these categories, no additional information about the shape of the object is taken into consideration, and hence, these techniques usually provide little improvement in the depth map. In this paper, we propose to incorporate a structural prior that helps to maintain the structural details in the recovered depth map. For this, we devise variations in the all-in-focus (AIF) image as the structural prior. By exploiting guided filtering, we improve the initial depth map through weighted least squares (WLS) based regularization for which our prior provides efficient weights. Experiments have been conducted on a variety of synthetic and real image sequences, and the results demonstrate that the proposed structural prior improves the accuracy of depth reconstruction.