The Research Of No Reference Image Quality Assessment Algorithm Based On Statistical And Structural Features

Image quality assessment has important application value in the process of image acquisition,transmission and processing.Compared with subjective assessment methods,objective assessment methods are divided into three categories: full reference,reduce-reference and no-reference.The first two methods need to be compared with the reference image when estimating the quality of the image,which leads to their limited application scenarios.No-reference method can directly estimate the image quality from the distorted image,which is more practical.In this paper,starting from the gradient,texture and multiple structure of the image,the no-reference image quality assessment method is researched.The main contents are as follows.(1)In this paper,a no-reference assessment method called Blind Gradient and Texture perception model(BGT)is proposed.Referring to the biological sensitivity of human visual system to image gradient and texture information,aiming at the problem that the existing no-reference methods do not fully consider the relationship between image gradient and texture,we propose to adopt conditional probability to model the relationship between gradient and texture,and apply the relationship to the task of image quality assessment.The local details of the image are captured by the local binary pattern operator,and the quality perception features are extracted in the form of histogram.Finally,the effectiveness of the BGT method in precision and generalization ability is tested on several data sets.(2)In this paper,an unsupervised assessment method called Rich Structural information-driven Natural Image Quality Evaluator(RS-NIQE)is proposed.Firstly,the coarse edge structure,fine texture structure and high-level visual structure of the image are extracted to sense the image distortion.Then,the fitting features,LBP histogram features and SVD features are extracted from the three feature maps respectively to make full use of the image structure information to describe the image quality change.The experimental results show that the three kinds of structural information are complementary in the task of image quality evaluation,and the proposed RS-NIQE model is superior to the existing unsupervised methods in prediction accuracy and generalization ability.(3)In order to facilitate users to evaluate image quality and develop new image quality evaluation algorithm,this paper develops an image quality evaluation software.We integrate the two algorithms and build an image quality evaluation system,which can easily use different models to estimate the image quality.In addition,we test the stability of the system and the function of each module.On the interface of the software,we can clearly see the image to be evaluated,image quality score and algorithm processing time.