Research On Evaluation Method Of No-reference Image Quality Based On Visual Attention Mechanism

Images are one of the important ways for people to perceive the objective world.So the clearer the image contains more information,the better the quality.However,images are often degraded or distorted due to various reasons during image acquisition and processing.To accurately assess the degree of image distortion and facilitate subsequent processing such as image enhancement and image compression,research on objective image quality evaluation has become increasingly important.Through careful study of the current objective image quality evaluation,it is found that most of the non-reference IQA methods ignore the application of the human visual attention mechanism,and the visual attention mechanism can theoretically improve the quality evaluation performance.Therefore,this article focuses on the characteristics of the visual attention mechanism and its application in the objective IQA field and gives two non-reference IQA methods from the perspective of visual attention.Firstly,this thesis is based on the CEIQ method and proposes an IQA method for contrast distortion.The factors affecting contrast are mainly considered from the perspective of visual attention.The first is from the statistical aspects of brightness information.The extracted features include CEIQ original features,standard deviation,histogram energy,and skewness.The second is to consider the influence of image color.The extracted features include color saturation and color saturation.Then use the extracted feature vector as the input of the SVR model to learn the predicted image quality score.Finally,verify and analyze the improved model in the CSIQ,CID2013,and CCID2014 databases.The results show that,compared with the IQA specifically used for contrast changes,the model in this thesis has achieved relatively good results,and the model has good generalization ability and stability.Secondly,this thesis improves a general-purpose non-reference IQA method combined with the visual attention mechanism.First,the ITTI model is used to divide the image to be tested into regions of interest and non-interests,and then the natural scene statistical features of the two regions are extracted through the ILNIQE model,and their corresponding MVG models are calculated according to these statistical features.The distance(quality score)of the MVG model of the two regions and the MVG model of the original natural image is weighted and fused to obtain the overall score of the image.Finally,the improved model in this thesis is tested in the CSIQ,LIVE,and TID2013 databases.The results show that the model in this thesis has a high performance in evaluating image quality,and its evaluation results are closer to human subjective feelings than the traditional general-purpose IQA model without reference.