Edge-Preserving Stereo Matching Using Minimum Spanning Tree

Congxuan Zhang,Wen Liu,Ming Li,Junjie Wu,Chao He,Zhen Chen

doi:10.1109/access.2019.2958527

Abstract

Despite that the accuracy and efficiency of stereo matching technology have significantly improved in the past decades, the issue of edge-blurring remains a challenge to most of the existing approaches. In this paper, we propose a minimum spanning tree (MST) based stereo matching method by using the image edge and segmentation optimization to preserve the image boundary. We first exploit a fast disparity range estimation method by combining the Surf and Akaze feature points to improve the computational efficiency. Second, we utilize the image edges and brightness information to generate a self-adaptive weight function, which is able to significantly improve the accuracy of MST aggregating in the regions of complex texture and boundaries with similar color distribution. Third, we employ the image segmentation to extract the invalid regions of the estimated disparity map, and propose a post-processing scheme to refine the disparity result. Finally, we run our method on several Middlebury and KITTI datasets. The comparison results between our method and other state-of-the-art approaches demonstrate that the proposed method has high accuracy for disparity computation and is especially robust to the edge-blurring.

Highlights

It is undoubted that the binocular stereo matching is a fundamental issue in the areas of image processing and computer vision
We propose a minimum spanning tree (MST) based cost aggregation approach by incorporating image edges with brightness information to define a reasonable weight of neighboring pixels
The experimental results of different modeling choices indicate that the proposed disparity range estimation, image edge constraint and image segmentation optimization can increase both the computing accuracy and efficiency of stereo matching, and the presented method is robust to various image sizes

Summary

INTRODUCTION

It is undoubted that the binocular stereo matching is a fundamental issue in the areas of image processing and computer vision. Still several noticeable challenges such as illumination variation, shape deformation, untextured regions and complex edges As a result, these widespread difficulties may lead to obvious errors generated by current stereo methods. To address the issue of edge-blurring, we propose here a minimum spanning tree (MST) based stereo matching method by using the image edge and segmentation optimization approach. In order to improve computation efficiency of stereo matching, we revised a disparity range estimation framework by combining the Surf and Akaze feature correspondence methods. To ensure the edge-preserving performance in the untextured regions, we explored a post-processing refinement approach by using image segmentation to recover the invalid regions of the computed disparity map.

RELATED WORKS

DISCUSSION OF FREE PARAMETERS

Findings

CONCLUSION