Blind image quality assessment with semi-supervised learning and fuzzy logic

Abstract

Blind image quality assessment (BIQA) is a challenging task due to the difficulties in extracting quality-aware features and modeling the relationship between the features and the visual quality. In this paper, we propose a semi-supervised and fuzzy (S2F) framework for BIQA. First, we formulate the fuzzy process of subjective quality assessment by using fuzzy logic. Secondly, we introduce the semi-supervised local linear embedding (SS-LLE) to learn the mapping from the features to the truth values using both the labeled and unlabeled images. Experimental results on two benchmarking databases demonstrate the effectiveness and promising performance of the proposed S2F framework for BIQA. © 2013 Springer-Verlag Berlin Heidelberg.