Research On Objective Image Quality Assessment

Image Quality Assessment(IQA)technology has always been one of the research hotspots in the field of computer vision.Image Quality Assessment aims to assess the degree of distortion of an image.Image quality not only affects people's visual experience,but also affects the performance of other image processing algorithms.In recent years,objective image quality evaluation algorithms have attracted the attention of a large number of researchers for their advantages in both speed and accuracy.At the same time,this type of algorithm also faces many challenges.When the image suffers from complex distortion types,it is prone to mismatch between the evaluation results and the subjective evaluation results.In order to further improve the accuracy of the algorithm under different challenge factors,this article mainly focuses on the objective full-reference image quality evaluation algorithm.The specific research content is as follows:1.An image quality assessment algorithm that integrates color characteristics is proposed.Traditional quality evaluation algorithms usually only use hue features when evaluating image color distortion.However,this feature is difficult to express complex color features,which leads to the problem of inaccurate evaluation of complex color distortion.In addition,the traditional saliency algorithm does not fully consider the color temperature tendency characteristics of the human eye,which leads to the problem that the obtained saliency map does not fit the visual characteristics of the human eye.In response to the above two problems,this article deeply analyzes the role of color characteristics in image quality evaluation,and makes the following improvements on the basis of the visual saliency index(VSI)algorithm: First,on the basis of hue evaluation,this paper introduced color vividness and depth features to express complex color features such as color saturation,thereby improving the accuracy of the algorithm for evaluating complex color distortion.Secondly,optimize the color temperature saliency map of the SDSP saliency algorithm,strengthen the influence of high color temperature areas on the saliency map,which improves the sensitivity of the saliency algorithm to colors.Finally,the proposed method is compared with other excellent image quality evaluation algorithms on TID2008 and TID2013 datasets.The experimental results show that compared with VSI,the accuracy of the proposed algorithm improves 19.49% on the Block distortion type and 3.34% on the MS distortion type.The SROCC values of the proposed algorithm in the two data sets reached 0.8993 and0.8994,respectively.2.An image quality evaluation algorithm based on adaptive weights between features is proposed.Traditional image quality evaluation algorithms often use the similarity calculation formula in the SSIM algorithm when calculating feature similarity.A large number of experiments have shown that this method is prone to omit complex features distortions.In addition,when calculating local image quality scores,traditional algorithms often use fixed inter-feature weights for feature fusion,ignoring the characteristics of the image itself,and easily lead to the problem of the main features of the image being not prominent.In order to solve the above problems,this article has made the following improvements on the basis of the image quality evaluation algorithm based on comprehensive color characteristics: Firstly,the traditional feature similarity calculation formula is improved.The similarity difference between the auxiliary image and the reference image and the distorted image is used to further highlight the difference between the distorted image and the reference image.Secondly,a set of calculation rules for the contribution degree of image features are proposed.The feature weights during feature fusion are confirmed according to the proportion of the feature contribution degrees,so that the weights between the features can be adjusted adaptively according to the image characteristics.The experimental results show that on the TID2008 and TID2013 data sets,the algorithm stability of the proposed algorithm is improved by 2.25% and 2.62% compared with the traditional algorithm,and the SROCC values of the proposed algorithm in the two data sets reach0.8996 and 0.9003 respectively.