Color Image Quality Assessment Based On Deep Learning

In recent years,image quality assessment has become a research hotspot in the field of image processing and computer vision.Among them,because the no-reference image quality assessment algorithm does not need to refer to the original image information,it is more applicable in practice,so it has become the main research directions in image quality assessment domain.When most traditional no-reference quality assessment algorithms evaluate the quality of color images,they first convert the color image into a grayscale image or into a certain color space to extract image information,and then use the average quality evaluation score of each channel of the image as the final quality evaluation score of the color image,which not only fails to fully extract the features of the color image,but also does not take into account the difference in the degree of attention of the human eyes to different areas of the image,resulting in poor subjective and objective consistency.It is quite necessary to explore the influence of the image feature extraction and human visual perception on color image quality assessment.The main research work and contributions of this paper are as follows:(1)Aiming at the problem that the current algorithm does not fully extract color image features when evaluating color image quality,a model that containing different color spaces input is proposed to extract image information from different color spaces of color images to analyze and study the characteristics the influence of image information on image quality assessment of different color spaces.Experimental results show that incorporating the image features extracted from color images of different color spaces can enable the model to obtain more color information.Considering that the features of different scales of the image can describe the information of different scales in the image,a multi-scale model is proposed based on this.By adding branches with different convolutional layers in the model,we can extract image features at different scales.Experiments results show that extracting image features of different scales can effectively improve the accuracy of the algorithm for predicting image quality,and provide theoretical supports for the color image quality assessment method based on deep learning proposed later.(2)Aiming at the problem that the current image quality assessment algorithm cannot well simulate the characteristics of human visual perception,a color image quality assessment model based on the visual attention mechanism is proposed.The features extracted from different color spaces in the color image are inputted into the model at the same time,multi-scale module is incorporated in the model to extract the features of different scales of the image,according to the difference of perception extent of different areas of the image by human eyes,the visual attention mechanism is introduced into the model,so that when calculating the final quality score in the image,different weights are assigned to the local quality scores of different regions in the image.The assessment model simulates the process of human visual perception system when observing color images.Experiments results show that the image quality assessment algorithm proposed in this paper has consistency with the subjective evaluation of human eyes.