No-reference Image Quality Assessment And The Application

In the current era of the big information explosion,digital images become the main carrier to transfer information.Human social activities produce a mass of image data.High quality and clear images bring a lot of convenience to people's social life.However,due to the current development level of information technology,various kinds of noise will be introduced into the course of image processing,such as generation,transmission and storage,which will result in the attenuation of image quality and directly affects people's acquisition of image information.How to measure this attenuation at the algorithmic level is of great practical significance.The study of the field of objective image quality assessment is the solution to this problem.Especially,the No-reference image quality assessment which has the fewest limits is widely applied in image compression,video surveillance and aerial UAV imaging system.In order to solve the problem of accuracy,efficiency and robustness of the No-reference image quality assessment algorithm,this paper proposes three kinds of NR-IQA algorithm based on the mechanism of human vision induction.First of all,cognitive neuroscience research indicates that natural images possess sparse structure.Inspired by this theory,we use dictionary learning method to establish the image feature space codebooks to represent image visual contents,in order to complete image quality prediction,sparse representation method is applied to feature pooling on the feature space codebooks.Based on this,a new NR-IQA method building on multilevel codebooks is proposed,which is used in the airborne imaging system;Secondly,Inspired by the human visual system of cortical orientation selectivity principle we design a new local image descriptor in the spatial domain to extract image visual patterns.Then,we establish the pattern dictionary based on a large number of visual patterns to extract the image feature vectors,which are used to capture image quality degradation caused by different kinds of noises.Afterwards,a novel NR-IQA method based on HVS orientation selectivity mechanism is proposed which is integrated in the embedded development board as a portable and real-time objective image quality evaluation system.Finally,from the view of Shannon information,information entropy is used to measure the degradation of information that an image contains.Researches on the visual cortex indicates that the local receptive field extract the image brightness,direction and gradient information to perceive image content.Inspired by this mechanism,we analyze these three kinds of image features' joint entropy,joint probability distribution,conditional probability distribution and marginal probability distribution to extract image feature vectors and establish regression model.Afterwards,an original NR-IQA method based on multi-class feature entropy degradation is proposed which can be used in video quality assessment system.