DocumentCode
257958
Title
COHERENSi: A new full-reference IQA index using error spectrum chaos
Author
Hegazy, Tamir ; Al Regib, Ghassan
Author_Institution
Center for Signal & Inf. Process. (CSIP), Georgia Inst. of Technol., Atlanta, GA, USA
fYear
2014
fDate
3-5 Dec. 2014
Firstpage
965
Lastpage
969
Abstract
We introduce a new family of quality indices that are based on the reported findings that biological visual systems exhibit interference in spatial harmonic distortion, and that local phase distortion is more perceived than global phase distortion. This paper specifically addresses a full-reference version of the proposed indices. The proposed quality index has three components: harmonic chaos analysis, phase chaos analysis, and pooling. The first two parts produce two maps that are weighted and summed to produce a single map. Then, we use the mean of the logarithmically scaled combined map as the quality index. We further propose a multi-scale variation of the index. Experimental evaluations were conducted on more than 1,300 images from the TID2 013 image database. We specifically test the quality index on three distortion categories: Simple, Actual, and Noise. Then, we cross-validate the results with the LIVE database. The results show that the proposed index outperforms all known indices based on correlation with subjective evaluation.
Keywords
chaos; distortion; harmonic analysis; image processing; visual databases; COHERENSI; TID2 013 image database; biological visual system; error spectrum chaos; full-reference IQA index; harmonic chaos analysis; image quality assessment; local phase distortion; phase chaos analysis; spatial harmonic distortion; Chaos; Discrete Fourier transforms; Harmonic analysis; Image quality; Indexes; Noise; Sensitivity; Full-Reference; Image Quality; Perception;
fLanguage
English
Publisher
ieee
Conference_Titel
Signal and Information Processing (GlobalSIP), 2014 IEEE Global Conference on
Conference_Location
Atlanta, GA
Type
conf
DOI
10.1109/GlobalSIP.2014.7032264
Filename
7032264
Link To Document