Image quality assessment using singular vectors

Chin Ann Yang, Mostafa Kaveh

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

3 Scopus citations

Abstract

A new Full-Reference Singular Value Decomposition (SVD) based Image Quality Measurement (IQM) is proposed in this paper. Most of the recently developed IQMs that have been designed for measuring universal distortion types have worse results in measuring blur type distortions. The proposed method A-SVD aims at capturing the loss of structural content instead of measuring the distortion of pixel intensity value. A-SVD uses the change in the angle between the principal singular vectors as a distance between the original and distorted image blocks. Experiments were conducted using the LIVE database. The proposed algorithm was compared with another recently proposed SVD based method named M-SVD and other well-established methods including SSIM, MSSIM, and VSNR. Results have shown that the proposed method has an advantage in discerning blurry types of image distortions while providing comparable results for other distortion types. Also, the proposed method provides better linear correlation with the human score, which is a desirable attribute for the IQM to be used in other applications.

Original languageEnglish (US)
Title of host publicationProceedings of SPIE-IS and T Electronic Imaging - Image Quality and System Performance VII
DOIs
StatePublished - 2010
EventImage Quality and System Performance VII - San Jose, CA, United States
Duration: Jan 18 2010Jan 19 2010

Publication series

NameProceedings of SPIE - The International Society for Optical Engineering
Volume7529
ISSN (Print)0277-786X

Conference

ConferenceImage Quality and System Performance VII
Country/TerritoryUnited States
CitySan Jose, CA
Period1/18/101/19/10

Keywords

  • Blur
  • Image quality measurements
  • Singular value decomposition (SVD)

Fingerprint

Dive into the research topics of 'Image quality assessment using singular vectors'. Together they form a unique fingerprint.

Cite this