Accurate Depth-from-Focus Reconstruction using Local and Nonlocal Smoothness Priors

摘要

In this paper, we tackle the problem of reconstructing a depth image from a focal stack, which is known as depth-from-focus (DFF) or shape-from-focus (SFF). Recovering a smooth depth image while preserving object structure is a typical issue associated with the conventional DFF or SFF reconstruction techniques. To address this issue, we propose a depth reconstruction method by including two existing local and nonlocal smoothness priors commonly used for natural image matting. Through the combination of local and nonlocal smoothness priors, we can reconstruct a depth image with sharp edges while maintaining spatial consistency. We demonstrate the effectiveness and robustness of the proposed algorithm over synthetic and real scene focal stacks in terms of accuracy and robustness compared to related approaches.