A Novel Depth Recovery Approach from Multi-View Stereo Based Focusing

Abstract

In this paper, we propose a novel depth recovery method from multi-view stereo based focusing. Inspired by the 4D light field theory, we discover the relationship between classical multi-view stereo (MVS) and depth from focus (DFF) methods and concern about different frequency distribution in 2D light field space. Then we propose a way to separate the depth recovery into two steps. At the first stage, we choose some depth candidates using existing multi-view stereo method. At the second phase, the depth from focusing algorithm is employed to determine the final depth. As well known, multi-view stereo and depth from focus need different kinds of input images, which can not be acquired at the same time by using traditional imaging system. We have addressed this issue by using a camera array system and synthetic aperture photography. Both multi-view images and distinct defocus blur images can be captured at the same time. Experimental results have shown that our proposed method can take advantages of MVS and DFF and the recovered depth is better than traditional methods.