LIGHT FIELD DEPTH ESTIMATION EXPLOITING LINEAR STRUCTURE IN EPI

Light field (LF), a promising representation to describe the visual appearance of a scene, implicitly captures 3D scene geometry. Inspired by this, we exploit the special linear structure of epipolar plane image (EPI) and propose a novel framework for depth estimation for 4D LF. Our approach estimates disparities through locating the optimal slope of each line segmentation on EPIs, which are projected by corresponding scene points. For each pixel to be processed, we employ intensity pixel value, gradient pixel value, spatial consistency as well as reliability measure to select the best slope from a predefined set. The depth value is calculated according to the geometric relation between disparity and slope of line segmentation in EPI. Then a novel method to detect and handle occlusion boundaries is introduced, further improving the quality of depth maps. We test our algorithm on not only a number of synthetic LF examples but real-world LF datasets, and the experimental results show that our technique outperforms both the state-of-the art and the recent light field stereo matching methods, especially near occlusion boundaries.