Disparity Refinement based on Depth Image Layers Separation for Stereo Matching Algorithms

Abstract

This paper presents a method to improve the raw disparity maps in the disparity refinement stage for stereo matching algorithm. The proposed algorithm will use the disparity depth map from the stereo matching algorithm as initial disparity depth output with a basic similarity metric of SAD. The similarity metric finds the pixel points between the left and right under the fixed window (FW) searching process. With this approach, the raw disparity depth map obtained is not smooth and contained errors particularly with the depth discontinuities and unable to detect the uniform areas and repetitive patterns. The initial disparity depth will be used to identify the layers of disparity depth map by adapting the Depth Image Layers Separation (DILS) algorithm that separate the layers of depth based on disparity range. Each particular disparity depth map distributed along the disparity range and can be divided into several layers. The layer will be mapped to segmented reference image to refine the disparity depth map. This method will be known as the Depth Layer Refinement (DLR) that using the disparity depth layers to refine the disparity map