Structured Computational Modeling of Human Visual System for No-reference Image Quality Assessment

Abstract Objective image quality assessment (IQA) plays an important role in various visual communication systems, which can automatically and efficiently predict the perceived of images. The human eye is ultimate evaluator for experience, thus modeling system (HVS) a core issue objective IQA experience optimization. traditional model based on black box fitting has low interpretability it difficult to guide optimization effectively, while physiological simulation hard integrate into practical services due its high computational complexity. For bridging gap between signal distortion this paper, we propose novel perceptual no-reference (NR) algorithm structural HVS. According mechanism brain, divide processing low-level layer, middle-level layer high-level conduct pixel information processing, primitive global respectively. natural scene statistics (NSS) features, deep features free-energy are extracted from these three layers. support vector regression (SVR) employed aggregate final prediction. Extensive experimental comparisons widely used benchmark databases (LIVE, CSIQ TID2013) demonstrate that our proposed metric highly competitive with or outperforms state-of-the-art NR measures.