Image Quality Assessment Method Based On Multiple Feature

With the intelligent development of society,image data has been used more and more widely in various fields,and people have higher and higher requirements for image quality.However,in the process of image acquisition,transmission and storage,image degradation and distortion are often caused due to various reasons.Image quality assessment(IQA)as an effective means of evaluating image quality has been widely concerned by many researchers.Because subjective image quality evaluation methods require the direct participation of human beings,they are not only time-consuming and labor-intensive,but also difficult to implement.Therefore,objective image quality evaluation methods have been widely studied.Because the full-reference objective image quality evaluation method can obtain all the information of the reference image,its evaluation result is relatively reliable,but some reference images are difficult to obtain,so the no-reference objective image quality evaluation method that gets rid of the dependence on the reference image becomes the first choice.Based on the characteristics of human vision,this paper studies a full-reference image quality evaluation method and a no-reference image quality evaluation method,and conducts experiments on the public LIVE database and the self-built image quality evaluation database.At present,most of the full-reference image quality evaluation methods extract a single feature of the image,and then calculate the quality score according to the constructed mathematical model.The value interval is often inconsistent with the value interval of the average opinion score.In order to facilitate the analysis of the results to make the value interval consistent,this paper takes the extracted image features as the input,and the corresponding average opinion score as the output.Through support vector regression learning,a full-reference image quality evaluation model is obtained,and the value of the evaluation result is the interval of the average opinion score is consistent.The full-reference image quality evaluation method based on image similarity features proposed in this paper extracts three similarity features of the image.On the basis of the single-feature training model,it also extracts multiple feature training models at the same time.The verification found that the evaluation result of the model trained with multiple features is better than asingle feature,and the performance indicators of the full reference method in this paper are better than the classic full reference methods PSNR,SSIM and GSSIM.Since the no-reference image quality evaluation method cannot obtain any information of the reference image,it can only be used for analysis by extracting the characteristics of the distorted image,so the learning-based method has become the first choice for the no-reference image quality evaluation.The existing no-reference image quality evaluation Most of the methods are to extract a single feature of the image to train the non-reference image quality evaluation model.The no-reference image quality evaluation method based on multiple image features proposed in this paper extracts two image features.Based on a single feature,two are also extracted jointly.The feature is used to train the support vector regression model.It is found on the test set that the evaluation result of the model trained with the two features is better than the single feature,and compared to the classic full-reference image quality evaluation methods PSNR and SSIM and the classic no-reference image quality The evaluation method BRISQUE has obtained better results.