A New Segment-Based Structure of Multistage Median Filter Algorithmfor Stereo Matching

Abstract

Stereo matching algorithms developmentgenerally focus on eliminating the undesired aspects includingocclusions, noises, horizontal stripes,and unwanted regions. In this paper, we present a Multistage Hybrid Median Filter (MHMF)method with a new segment-based of a stereo matchingalgorithm as the main core of our research approach.MHMF is a new hybrid algorithmwith a high ability to obtainan accuratedisparity depth map that embedded into numerous stereo matching applications. The Multistage Hybrid Median Filter (MHMF)method typically involvesseveral approaches such as Basic Block Matching (BBM), Dynamic Programming (DP), and Hybrid Median Filter (MHF).The significant contribution of MHMFconcerns on segmentation and merging processes. Those processesare developed to solve multiple issues and drawbacks including depth map non-edge preserving, horizontal stripesof the raw disparity depth map, noises, and other unresolved challenges that are based on the presence of undesired aspects and regions. The segmentation process of MHMF is to segment the optimized raw disparity map into several parts based on the pixel colors. The merging approach is mainly to combine allsegmented segments into a disparity map.The datasets used for the MHMF algorithm are taken from Middlebury Stereo Vision Page and captured by MVBLUEFOX stereo camera. Eventually, this paper will provide a computation of MHMF, where various functions are used to evaluate the obtained results includingthe system of Middlebury, Similarity Structural Index Metric (SSIM), Peak to Signal Noise Ratio (PSNR) and Mean Square Errors (MSE).
Experimental results show thatthe proposed method achieves high results compared to many existing algorithms.

 Keywords: Stereo matching algorithm, Segment-based, Multistagehybrid filter, Mergingalgorithm.