Multistage Hybrid Median Filter (MHMF) Design for Stereo Matching Algorithms

Abstract

In this paper, a Multistage Hybrid Median Filter (MHMF) of the stereo matching algorithm is proposed with high capability of providing a high performance and high accuracy of disparity depth map to be embedded in numerous stereo matching applications. The significant contribution of this approach is to solve multiple drawbacks and inherent issues that make stereo difficult and unresolved including noises, horizontal stripes, depth map non-edge preserving, and removing the unwanted regions that affect the quality of the disparity depth map. A new post-processing algorithm is developed which occupied multiple stages including a new segment-based, multiple filtering processes, and merging approaches in order to achieve our targets. The structure of MHMF is consisting of several parts including Basic Block Matching (BBM), Dynamic Programming (DP), segmentation, Hybrid Median Filtering (MHF), and merging processes toward the achievement of high disparity depth map. The paper will also provide the performance of MHMF with some existing algorithms, which they all will be evaluated using several stereo functions including PSNR, MSE, and SSIM. Based on the evaluation, the proposed approach can achieve high accuracy of disparity depth map with less algorithm complexity and computational efficiency among other algorithms. Thus, the Multistage Hybrid Median Filter (MHMF) can be a unique approach for efficient performance of stereo vision and 3D applications.