No Reference Image Quality Assessment And Optimization Based On Visual Perception

With the coming of information age and the rapid development of multimedia and internet technology,people are more and more dependent on the visual information.Due to the process of acquisition,transmission,images are easily to suffer with various types of distortion.In the real life it is difficult to obtain the original reference image,which seriously influences the effective use of images.Thus,the research on no reference image quality assessment plays a very important role and has lots of academic value as well as application value.For the problem that the most current quality assessment methods of images with single distortion can not accurately reflect the effect of human vision system,the thesis proposed an effect metric based on local information statistics of natural images.In the proposed method,discrete wavelet transform(DWT)is utilized to obtain the high frequency information of images.The DWT coefficients are divided into blocks at size of 8?8,then the mean values and skewness values are respectively calculated from the amplitudes of each block.Discrete cosine transform(DCT)is also utilized to get the frequency information of images which are divided into blocks at size of 8?8 in the spatial domain first.Then the mean values and skewness values are also respectively calculated from the amplitudes of each DCT block.The features are trained and tested by Support Vector Machine(SVM)to build the quality assessment model.The results of experiments on LIVE database show that the performance of proposed method is better than the most present classical methods.And also the proposed method is well consistent with the subjective assessment results,and can reflect human subjective feeling very well.Aiming at the problem that existing image quality assessment methods for single distortion perform poor in the assessment of multiply-distorted images,this thesis proposed a method which is based on edge structure and brightness information statistics.In this metric,Phase Congruency model and Local Normalization model are used to obtain the structural information and luminance information of images.Then,the Local Binary Pattern(LBP)is adopt to code these information and the probability histogram of every pixel is extracted to capture the changes of pixel caused by distortions.SVM is also used to model the relationship between the extracted features and the objective DMOS values of images.Experiments on MLIVE database andMDID2013 database verify that the proposed method is High consistency with the human vision perception.This thesis proposed a system for multiply-distorted images quality assessment and optimization which is combined with distortion classification and decoupling control.In this system,effect image features are first extracted to build the distortion classification and quality assessment model.Then,setting the score threshold to determine whether the image quality needs to be improved.In the process of image restoration in order to eliminate the coupling effect between different distortions,the decoupling model in the control theory is introduced into the system which ensures that some certain type distortion optimization algorithm would not affect the other distortions,avoiding the secondary pollution.