A no-reference image quality assessment approach based on steerable pyramid decomposition using natural scene statistics

For majorities of no-reference image quality assessment (NRIQA) algorithms, prior knowledge about image distortion types is needed. However, the source distortions in the image are actually not given in most cases. In this paper, we propose a no-reference image quality assessment approach based on steerable pyramid decomposition using natural scene statistics without any prior knowledge about the distortions of the original image. Because the means of (\(\log _{2}\) of) subband coefficient amplitudes (MLSCAs) of the natural images have certain statistical properties independent of their contents, a predicted MLSCAs based on these properties can be considered as a reference index for assessment. A subband distortion is then defined as the difference between the reference and the real MLSCA of the distorted image. As against most NRIQA algorithms that are distortion specific, the subband distortion determined is independent of distortion types. Therefore, the proposed method is capable of assessing the quality of a distorted image across multiple distortion categories without any prior knowledge about the distortions of the original image. Finally, a set of weights for each subband, trained from the subjective mean opinion scores in the LIVE image database, is used to combine the subband distortions into a quality score for evaluating the distorted images. Experimental results show that the proposed method outperforms five no-reference algorithms using natural scene statistics.