A Multistage Hybrid Median Filter Design of Stereo Matching Algorithms on Image Processing

Abstract

This paper presents a new method of stereo matching algorithms known as a Multistage Hybrid Median Filter (MHMF). The main challenging problem within computer vision field is stereo matching, considering the many drawbacks and issues resulting from stereo matching, which makes it a difficult and unresolved approach. Large studies and methods have been introduced and attempted to deal with stereo matching problems from a different perspective. However, most of these existing methods still suffer from low accuracy and algorithms complexity. Thus, to provide valuable and efficient solutions to stereovision community and to support researchers in the same line of research, we created a highly efficient and robust algorithms based on Hybrid Median Filter (MHF). In this paper, the (MHMF) is introduced as a newly implemented method of stereo matching algorithms. This method consists of three main stages, in which each stage involves multiple processes. In stage 1, the Basic Block Matching (BBM), Dynamic Programming (DP) algorithms are accomplished. While stages 2 and 3 rely on newly developed post-processing algorithms, which involve Hybrid Median Filtering (MHF), segmentation and merging processes as the main core of our research approach. The significant contribution of this method is on its capability of solving major drawbacks and problems of stereo marching including the high noises, horizontal stripes, multiple unwanted regions and aspects, and occlusions problems. The paper provides an evaluation of our method with some existing algorithms using the most common stereo functions including the MSE, PSNR, and SSIM.  Finally, it is found that our developed algorithms achieved high accuracy of disparity depth map with convenient execution time among other compared methods.