A Novel Image Quality Assessment Metric based on Image Gradient Vectors

Abstract

Image degradation deforms the structure of image, thus analysis of structural distortions of image is beneficial to image quality assessment (IQA). This paper proposes a novel structural information-based IQA metric, in which image gradient vector, composed of magnitude and phase components, is used to describe the structural information of an image. Since the phase quantization code of the phase component can also reflect the change of image quality to some extent, Hamming distance between the phase quantization code of the reference and distorted images is defined to represent the quality map of an image for quality assessment. Because different regions in image are with different importance for human visual perception, threecomponent image model for IQA is exploited according to the gradient magnitude component, and nonuniform weights are designed to emphasize the importance of different regions in image. Experimental results of the LIVE database show that the proposed IQA metric is more consistent with subjective results, and has higher correlation coefficients compared with other classic IQA metrics.