Depth and All-in-Focus Image Estimation in Synthetic Aperture Integral Imaging Under Partial Occlusions

Abstract

A common assumption in the integral imaging reconstruction is that a pixel will be photo-consistent if all viewpoints observed by the different cameras converge at a single point when focusing at the proper depth. However, the presence of occlusions between objects in the scene prevents this from being fulfilled. In this paper, a novel depth and all-in focus image estimation method is presented, based on a photo-consistency measure that uses the median criterion in relation to the elemental images. The interest of this approach is to find a solution to detect which camera correctly sees the partially occluded object at a certain depth and allows for a precise solution to the object depth. In addition, a robust solution is proposed to detect the boundary limits between partially occluded objects, which are subsequently used during the regularization depth estimation process. The experimental results show that the proposed method outperforms other state-of-the-art depth estimation methods in a synthetic aperture integral imaging framework.