DocumentCode :
191054
Title :
An adaptive window stereo matching based on seed voting
Author :
Bo Liu ; Qian Liang ; Yingyun Yang
Author_Institution :
Satellite Monitoring Dept., State Radio Monitoring Center, Beijing, China
fYear :
2014
fDate :
5-8 Aug. 2014
Firstpage :
771
Lastpage :
774
Abstract :
Stereo matching remains a difficult vision problem for the image noise, textureless regions, depth discontinuities and occlusions. This paper presents a novel adaptive window stereo matching algorithm based on seed voting to solve these ill-posed problems. There are three key contributions in this paper: combining color and gradient as conditional tags to restrain initial window size for further adaptively updating window shape; in terms of disparity optimization, improving the original Left-Right-Difference method to achieve more precise initial seed map; proposing an efficient seed growth based on vote for finial dense disparity map. Experimental results based on Middlebury stereo dataset demonstrate the superior performance of the proposed algorithm. The novel local-based approach stimulates stereo matching research to further development since many advanced stereo algorithms call for accuracy local results as initial disparity map.
Keywords :
computer vision; image matching; optimisation; stereo image processing; Middlebury stereo dataset; adaptive window stereo matching algorithm; conditional tags; dense disparity map; depth discontinuities; disparity optimization; image noise; left-right-difference method; local-based approach; occlusions; seed voting; textureless regions; vision problem; Algorithm design and analysis; Conferences; Image color analysis; Optimization; Shape; Stereo vision; Windows; Left-Right-Difference; Seed growth; Stereo matching;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Signal Processing, Communications and Computing (ICSPCC), 2014 IEEE International Conference on
Conference_Location :
Guilin
Print_ISBN :
978-1-4799-5272-4
Type :
conf
DOI :
10.1109/ICSPCC.2014.6986301
Filename :
6986301
Link To Document :
بازگشت