Initial depth estimation using adaptive window size with light field image

Abstract

Light field image generates a set of focusing images at different depth levels, which suggests that one can determine object, foreground, or background regions. Moreover, the image can provide another set of multiple view images with slight shifts (sub-aperture images) which allows disparity computation without the image calibration and rectification steps. Nevertheless, the baseline between sub-aperture images is narrow. It limits the depth estimation range and accuracy. To solve this problem, we present an initial depth estimation method using adaptive window matching on epipolar plane images (EPIs) and structure tensor, which appropriately applied to the sub-aperture light field images. We analyze the sub-aperture images to EPIs and compute depth and corresponding information using the structure tensor on EPIs. An adaptive window size is used to improve the robustness of corresponding information. Finally, an initial depth map of light field image is reconstructed.