Objective Quality Assessment Of Screen Content Image Based On Multi-feature Fusion

With the rapid development of mobile terminal,Internet network information processing technology and network multimedia communication platform in China,the generation number and processing requirements of Screen Content Image(SCI)are increasing rapidly.However,problems of noise interference and image distortion are easily to be produced in the process of obtaining,transmitting and compressing SCI.The image quality degradation caused by such interference and distortion is difficult to be accurately assessed and predicted.The assessment model based on natural image cannot play an effective role in SCI visual quality prediction,which has become a key problem restricting the improvement of SCI assessment index.Thus,the only way to solve this problem is to analyze the quality assessment method of objective screen content image efficiently and establish SCI model.Based on this,this paper studies the quality assessment method of screen content image based on multi-feature fusion.The research process and conclusions can be found as follows:A full-reference SCI assessment method based on gradient feature and traditional feature fusion is proposed in this paper.The proposed method fully takes the characteristics of the screen content images into account.And aiming at the strong edge property of SCI,this method proposes to use gradient operator to extract edge information and texture information in spatial domain.At the same time,phase consistency is extracted in frequency domain.T'he gradient information and edge information extracted by phase are fused to enhance the structure information of screen content image.In addition,Gauss difference is used to extract corner features and spatial frequency features in the frequency domain to supplement the gradient and phase congruency features,which is used to characterize the perceived quality of the image.Finally,the method uses Random Forest(RF)as learning tool to learn the relationship between the quality of the screen content image and the subjective perception of the human eye,and uses the regression model which has finished learning to predict the visual quality score of the image.A non-reference SCI assessment model based on gradient feature and depth feature fusion is proposed in this paper.The model holds that deep learning can better simulate the human brain in interpreting image and text data.On the one hand,the model introduces the Resnet-50 neural network into the screen content image,because it can extract the deep information of the screen content image well.On the other hand,extracting the gradient feature of the screen content image can supplement the edge information of the depth feature loss.This method is not limited to distortion types of screen content images.The results show that the PLCC assessment indexes of the two methods proposed in this paper in the SIQAD database reach 0.9010 and 0.9556 respectively.Therefore,their prediction ability is significantly better than other mainstream assessment algorithms.