Modified Dual Winner Takes All Approach for Tri-Stereo Image Matching Using Disparity Space Images

In this research paper, a new framework is proposed to increase the total number of correct matches for stereo correspondence using tri-stereo images. The research work investigates some of the less explored properties of Disparity Space Image (DSI), and considers the local maxima in addition to the global maximum of the cost function and propose a new tri-stereo matching method as Modified Dual Winner Takes All (MDWTA) using edges of disparity space image. Conventionally using a single DSI from a stereo pair optimization techniques are applied to get the disparity, whereas the proposed approach uses Winner Takes All (WTA) approach using two DSIs from a triplet. The dual property of DSI that is the manner in which it stores cost for rows and columns for forward and reverse matching is introduced. This property is applied to check consistency for forward and reverse matching in a single pass, which gives initial correct matches. Next a left-centre-right consistency check is applied to discard inconsistent disparities obtained from these three views. Subsequently to obtain disparity for gaps thus generated a rule is formulated using local maxima, edges and adjacent global maximum to find correct disparity. Evaluation of the results obtained is carried by comparing them with block matching WTA method and other recent methods such as, Dynamic Programming (DP) and Semi-Global Matching (SGM) applied on images from Terrain Mapping Camera of Chandrayaan-1 and PRISM sensor of the ALOS mission. Methodology proposed is also verified using standard Middlebury stereo dataset. Experimental results show that the proposed method gives an increased number of correct matches by 10–15 % as compared to basic WTA and DP. However the results obtained are better to SGM for certain regions and features.