Improved stereo correlation using moravec operator and kolmogorov-smirnov test

Ananda S. Chowdhury, Hamid R. Arabnia, Dipankar Bandyopadhyay

Research output: Chapter in Book/Report/Conference proceedingConference contribution

2 Scopus citations

Abstract

Extraction of effective depth information from a stereo pair is a classical image processing problem. In the present paper, the performance of a mean intensity level invariance based template matching algorithm is improved using Moravec operator and Kolmogorov-Smirnov test. In the original template based stereo-correlation algorithm, the entire images are compared using correlation and search windows for the purpose of extracting useful depth information. However, it is generally observed that the entire stereo pair is not a good source of depth information. For the purpose of identifying potential areas containing depth information like the edges and corners, Moravec operator is employed followed by an iterative thresholding procedure. On the other hand, Kolmogorov-Smirnov test is applied to extract areas of comparable intensities. The incorporation of this statistical test thus reduces false comparisons between the two images forming a stereo pair at the time of identifying depth information. Furthermore, since the Kolmogorov-Smirnov test is a nonparametric hypothesis testing method, in order to apply the test, we don't need to assume any specific intensity distribution for the stereo pair. The performance of the stereo correlation algorithm is found to be improved with the incorporation of the above two tools/techniques. Experimental results on real stereo pairs are presented.

Original languageEnglish (US)
Title of host publicationProceedings of the 2005 International Conference on Computer Vision, VISION'05
Pages24-30
Number of pages7
StatePublished - Dec 1 2005
Event2005 International Conference on Computer Vision, VISION'05 - Las Vegas, NV, United States
Duration: Jun 20 2005Jun 23 2005

Publication series

NameProceedings of the 2005 International Conference on Computer Vision, VISION'05

Other

Other2005 International Conference on Computer Vision, VISION'05
CountryUnited States
CityLas Vegas, NV
Period6/20/056/23/05

Keywords

  • Kolmogorov-smirnov (KS) test
  • Moravec operator
  • Nonparametric statistics
  • Template matching

Fingerprint Dive into the research topics of 'Improved stereo correlation using moravec operator and kolmogorov-smirnov test'. Together they form a unique fingerprint.

Cite this