Fuzzy regression for perceptual image quality assessment. (August 2015)

• Record Type: Journal Article
• Authors: Chan, Kit Yan; Engelke, Ulrich
• Is Part Of: Engineering applications of artificial intelligence. Volume 43(2015:Jul.)
• Page Start: 102
• Page End: 110
• Publication Date: 2015-08
• Subjects: Fuzzy regression -- Mean opinion scores (MOS) -- Subjective image quality assessment -- Objective image quality metric; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2015.04.007
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Subjective image quality assessment (IQA) is fundamentally important in various image processing applications such as image/video compression and image reconstruction, since it directly indicates the actual human perception of an image. However, fuzziness due to human judgment is neglected in current methodologies for predicting subjective IQA, where the fuzziness indicates assessment uncertainty. In this article, we propose a fuzzy regression method that accounts for fuzziness introduced through human judgment and the limitations of widely-used psychometric quality scales. We demonstrate how fuzzy regression models provide fuzziness information regarding subjective IQA. We benchmark the fuzzy regression method against the commonly used explicit modeling method for subjective IQA namely statistical regression by considering three real situations involving subjective image quality experiments where: (a) the number of participants is insufficient; (b) an insufficient amount of data is used for modelling; and (c) variant fuzziness is caused by human judgment. Results indicate that fuzzy regression models achieve more effective data fitting and better generalization capability when predicting subjective IQA under different types and levels of image distortion.