Color Image Quality Evaluation based on Visual Saliency and Gradient Information

In the field of image quality evaluation, visual saliency and gradient information are very effective features for quality evaluation models. Visual saliency is often used to study which areas of an image are most attractive to the human visual system. Moreover, the degradation of gradient information can reflect the degree of structure distortion of images. Considering these two points, we propose a simple but very effective quality evaluation metric for color images. After obtaining the local gradient similarity information, the similarity of visual saliency and color information are also calculated, and then we calculate the standard deviations of the three components to obtain the final quality score. The experimental results from five benchmark databases (LIVE, IVC, TID2008, TID2013 and CSIQ) show that our model performs better than other methods in the correlation with human visual quality judgment.