Blind Image Quality Assessment Based on Local Quantized Pattern

Abstract

No-reference (NR) image quality assessment (IQA) metrics have attracted great attention in the area of image processing. Since there is no access to the reference images, the generic NR IQA metrics have made less progress than the full-reference and reduced-reference IQA metrics. In this paper, we aim to propose an effective quality-aware feature based on the local quantized pattern (LQP) for quality evaluation. Firstly, a codebook is learned by K-means clustering the LQP descriptors of a corpus of pristine images. Based on the codebook, the LQP descriptors of images are then encoded to derive the quality-aware features. Finally, the image features are mapped to the subjective quality scores using the support vector regression. Experimental results on several public databases indicate the propose method performs highly consistent with the human visual perception.