Abstract

In this paper, we propose a novel method for depth esti-mation in light fields which employs a specifically designedsparse decomposition to leverage the depth-orientation re-lationship on its epipolar plane images. The proposedmethod learns the structure of the central view and uses thisinformation to construct a light field dictionary for whichgroups of atoms correspond to unique disparities. This dic-tionary is then used to code a sparse representation of thelight field. Analyzing the coefficients of this representationwith respect to the disparities of their corresponding atomsyields an accurate and robust estimate of depth. In addi-tion, if the light field has multiple depth layers, such as forreflective or transparent surfaces, statistical analysis of thecoefficients can be employed to infer the respective depth of the superimposed layers.