Indonesian Journal of Electrical Engineering and Computer Science Vol. 32, No. 3, December 2023, pp. 1512~1520 ISSN: 2502-4752, DOI: 10.11591/ijeecs.v32.i3.pp1512-1520  1512 Journal homepage: http://ijeecs.iaescore.com A study of stereo matching algorithm on low texture and depth discontinuity regions Melvin Gan Yeou Wei 1,2 , Rostam Affendi Hamzah 1 , Nik Syahrim Nik Anwar 2 , Adi Irwan Herman 3 1 Department of Electronic Engineering Technology, Fakulti Teknologi Kejuruteraan Elektrik and Elektronik, Universiti Teknikal Malaysia Melaka, Melaka, Malaysia 2 Department of Electrical Engineering, Fakulti Kejuruteraan Elektrik, Universiti Teknikal Malaysia Melaka, Melaka, Malaysia 3 Product and Test Engineering, Texas Instruments, Melaka, Malaysia Article Info ABSTRACT Article history: Received Apr 19, 2023 Revised Jul 30, 2023 Accepted Sep 11, 2023 This article studies the performance of the proposed stereo matching algorithm on complex regions. These regions are areas with very limited information for the matching process which are low texture, and depth discontinuity regions. In this study, each algorithm uses different matching cost computation (MCC) techniques, but for cost aggregation (CA), disparity optimization (DO) and disparity refinement (DR), the technique remains the same. The MCC are absolute difference (AD), the combination of absolute difference and gradient matching (AD+GM) and census transform (CT). Then, for CA, DO and DR, they are minimum spanning tree (MST), winner take all (WTA) and bilateral filter (BF), respectively. The results are presented and discussed in this article. Hence, thru this study the robust method can be estimated at the MCC stage. Keywords: Census transform Computer vision Depth discontinuity region Low texture region Minimum spanning tree Stereo matching algorithm This is an open access article under the CC BY-SA license. Corresponding Author: Rostam Affendi Hamzah Department of Electronic Engineering Technology, Fakulti Teknologi Kejuruteraan Elektrik and Elektronik Universiti Teknikal Malysia Melaka Taman Tasik Utama, 75450 Ayer Keroh, Melaka, Malaysia Email: rostamaffendi@utem.edu.my 1. INTRODUCTION Generally, stereo matching algorithm consists of 3 structures which are local based algorithm, semi- global based algorithm and global based algorithm. Fundamentally, the basic stereo matching algorithm consists of 4 stages and they are matching cost computation, cost aggregation, disparity optimization and disparity refinement. By referring to the evaluation from [1], there are improved methods such as dynamic programming (DP), scanline optimization (SO), simulated annealing (SA) and graph cut (GC) which were introduced and replacing the winner takes all (WTA) approach. Global based often skip cost aggregation by defining global energy function. Yet, this approach remains unpopular due to complexity of the algorithms at that time. Then, through the comparative study by [2], the study reveals that adding the support weight (weight) for the support region (support window) on cost aggregation technique also produced an accurate disparity map. This technique is well-known among the researchers which this approach normalizes the support region weight with the designated or suitable energy on the pixel of interest. Finally, semi-global based algorithm, which proposed by [3] was introduced through the combination of local based technique and global based technique. The basic taxonomy of stereo vision disparity map (SVDM) algorithm mainly include 4 stages, yet one crucial stage always remains, matching cost computation (MCC). Here, left image known as reference image will correspondence with the right image; or the targeted image to produce the disparity map [4].