Image quality/distortion metric based on α-stable model similarity
Chongwu Tang, Xiaokang Yang, Guangtao Zhai · 2014
State-of-the-art objective image quality assessment (IQA) algorithms are generally designed for specific distortion types or based on complicated training procedures. For general-purpose applications, we propose a wavelet based no-reference IQA index nominated by Image Quality/Distortion Metric (IQDM). We characterize the leptokurtic and heavy-tailed behaviors of natural image's wavelet coefficients by symmetric α-stable (SαS) probability density, and distortion-caused distribution deviation is then used to gauge image quality by comparing degraded and reference SαS models in multi-scale wavelet sub-bands. We investigate the relationship between original and degraded parameters over scales, accordingly infer the reference parameters from the degraded ones using a linear approximation. A characteristic PDF for SαS is introduced and its closed form of Kullback-Leibler distance (KLD) is derived as IQDM using the model parameters. Extensive experiments demonstrate the highly comparable performance of IQDM to the distortion-sensitive or training based methods in consistency with subjective assessment scores.