An Algorithm for No-Reference Image Quality Assessment Based on Log-Derivative Statistics of Natural Scenes

摘要

In this paper, we propose a new method for blindo-reference image quality assessment based on the logderivative statistics of natural scenes. The new method, called DErivative Statistics-based Image QUality Evaluator (DESIQUE), extracts image quality-related statistical features at two image scales in both the spatial and frequency domains, upon which a two-stage framework is employed to evaluate image quality. In the spatial domain, normalized luminance values of an image are modeled in two ways: point-wise based statistics for single pixel values and pairwise-based log-derivative statistics for the relationship of pixel pairs. In the frequency domain, log-Gabor filters are used to extract the high frequency component of an image, which is also modeled by the log-derivative statistics. All of these statistics are characterized by a generalized Gaussian distribution model, the parameters of which form the underlying features of the proposed method. Experiment results show that DESIQUE not only leads to considerable performance improvements, but also maintains high computational efficiency.