Local Quality Pattern (LQP): A Quality Aware Local Statistical Feature Descriptor for No-Reference Image Quality Assessment

Assessment of the quality of the no-reference images in real-life applications is of immense importance. The no-reference image quality assessment (NR-IQA) algorithms are designed to measure quantitatively and accurately the perpetual quality of an image without any prior knowledge about its reference. In this work, we have proposed a novel NR-IQA metric to replicate the human visual system (HVS) for the perception of image distortion using the structural details and luminance information of the image. We propose a local descriptor that extracts the perceptual structural features of the image with highlighted reflectance to represent the variation of illuminance in the distorted probe image. We name this quality-aware descriptor as Local Quality Pattern (LQP). The extracted features are trained with support vector regression (SVR) to model the mapping of the feature vectors with the quality measure. Finally, the metric obtained is named Local Quality Pattern Score (LQPS). Extensive experiments conducted on publicly available datasets namely, LIVE, MLIVE and CSIQ databases to show that the proposed LQPS metric outperforms the compared NR-IQA methods with a prominent margin.