No-reference Image Quality Assessment Based On Natural Scene Statistics And Region Of Interest

The images that people get are inevitably degraded in quality for a variety of reasons.The no-reference image quality assessment(NR-IQA)method does not provide the original image as a reference when evaluating the image,so it has the advantage of directly evaluating the image and can be embedded in the real-time system to evaluate the image.Natural scene statistics(NSS)is the most essential attribute of the image,it is practical to extract it from the image as the feature to evaluate the image.Taking into account the attention of the human visual mechanism,the interest area will put more attention.Therefore,this paper based on the natural scene statistics and regions of interest(ROI),without the original image as a reference to the case of the image evaluation.The results of the paper are as follows:1.According to the natural scene statistics and the advantages of multi-scale analysis of image processing,this paper presents a NR-IQA method based on nonsubsampled contourlet transform(NSCT).This method analyzes the advantages of nonsubsampled contourlet transform,and then using it to extract the natural scene statistics of the image as features,and then uses the support vector regression(SVR)training model to obtain the image score.The validity of the proposed method is verified by comparing with the existing no-reference image quality assessment method on the LIVE data set.2.Considering the attention of the human visual mechanism,this paper presents a NR-IQA method based on the regions of interest.This method improves the Itti model by analyzing the perceived characteristics of the human eye,adding texture features and edge features.The saliency map extracted by the improved Itti model can better embody the interest area of the human eye.Then,the natural scene statistics of the image are extracted by using the nonsubsampled contourlet transform,and the distance between the regions of interest and the regions of non-interest to the original image is calculated respectively.Finally,the image quality score is obtained by assigning different weights to the regions of interest and the regions of non-interest.Experiments show that this method is better than the performance of the method proposed in(1),and it is similar to the advanced no-reference image quality assessment method.The most important thing is that the method does not need to be trained.