The Depth Estimation Method Based on Double-Cues Fusion for Light Field Images

Abstract

As an important attribute of spatial information, depth information is indispensable in many applications such as object tracking. Unlike the traditional imaging technology, which only records the position information of light propagation, the light field camera can record the position and direction information of light at the same time. By processing the light field image obtained from one shot, multiple images of different views and refocused images can be obtained, which provides parallax cue and ambiguity cue for depth estimation. Different from other researchers who only use a single cue in parallax or ambiguity cue, this paper adaptively fuses the two cues, constructs a unified combined calculation model, and achieves a double-cues fusion method for light field depth estimation. We have tested the algorithm on the public dataset of 4D light field benchmark, and achieved the top rank on the MSE metric.