Image quality assessment using regularity of color distribution

Abstract

To quantify the perceptual quality of the color image, a novel reduced-reference computational model is proposed in this paper. The proposed metric, named regularity of color distribution measure, is designed to measure the difference of the color distribution regularity between the reference image and the distorted one. In particular, three kinds of channels (bright channel, median channel, and dark channel) are first extracted from the color image. Then, a fractal analysis is employed to characterize each channel by fractal dimension. Finally, we make use of fractal dimensions extracted from channels to predict the image quality. The proposed metric is evaluated on three largest color image databases (TID2013, CSIQ, and LIVE databases) and experimental results indicate that our metric achieves excellent performance.