Disparity Estimation using Light Ray Pair in Stacked 3D Light Field

Abstract

Light field (LF) is a concept that defines a number of light rays passing through free space, and creates novel views and estimates three-dimensional (3D) information by combining some of the light rays. Among them, the stacked 3D LF is a structure made by stacking several 3D LFs, and it supports free viewpoint movements over a wide range. However, 3D LF is a relatively simple structure and its use is limited due to insufficient vertical light rays. This paper proposes an effective method for estimating vertical disparity between 3D LFs in a stacked 3D LF. This vertical disparity can be used as key information to release the limitation. The contribution of this paper is as follows. First, this paper defines light ray pair in a stacked 3D LF. Light rays corresponding to light ray pair share the same foreground and background, and there is only vertical disparity. Second, based on light ray pair, this paper proposes an effective image reconstruction for disparity estimation. This method reconstructs the reference image by connecting all light rays that are light ray pair with the target image from the neighboring 3D LF. Since this reference image has only vertical disparity, disparity estimation becomes relatively easy. Additionally, the image-based disparity estimation may damage the continuity of 3D LF, so post-processing is applied to compensate for this. In the experimental results, the proposed method shows a noticeable improvement in objects with large disparity that are close to the camera.