Fuzzy logic modeling for objective image quality assessment

Abstract

This paper presents a novel methodology of objective image quality assessment (IQA) based on Fuzzy Logic (FL) method. The main purpose is to automatically assess the quality of image in agreement with human visual perception. The used attributes (quality metrics) and evaluation criteria (human rating mean opinion score MOS) are extracted from image quality database TID2013. The fuzzy model design starts by selecting the most independent attributes, by applying Pearson's correlation approach and seeking the most correlated metrics with the corresponding MOS. Then, Adaptive Neuro-Fuzzy Inference System (ANFIS) is applied in order to construct an objective fuzzy model able to efficiently predict the image quality correlated with the subjective MOS. In this paper, different fuzzy models are produced by modifying certain ANFIS configurations. After that, we select the appropriate ANFIS model that provides high prediction accuracy and stability with taking into account its implementation complexity. The overall architecture of the selected FL model consists of four input metrics, two bell-shaped membership functions associated to each input metric, two fuzzy if-then rules, two linear combination equations and one output which gives the image adequate quality score. Finally the performance of the proposed fuzzy model is compared with other IQA models produced by different machine learning methods, the simulation results demonstrate that the fuzzy logic model has a high stable behavior with the best agreement with human visual perception.