Image Quality Assessment And Its Application

In recent years,with the emergence of new media forms such as high-definition television,Internet video streaming and video on demand,digital visual information such as image and video is becoming more and more important in daily life and work.Whether it is image or video,the image quality will cause quality loss in the whole process of generation and propagation.Therefore,how to evaluate the received image quality plays an important role in image and video processing and communication.At the same time,image quality assessment(IQA)algorithm has attracted researchers' attention.With the continuous development and maturity of 3D technology,stereo image is widely used in various fields because of its advantages in reality.In this paper,first of all,starting from the research of stereo image quality,making full use of the viewer's perception of quality,combining with the extraction of multiple feature indexes of the whole stereo image,a method of no reference stereo image quality assessment(NR-SIQA)based on visual salience and perception is proposed.In the field of marine plankton research,image quality has a great impact on the task of positioning,tracking and classification of marine plankton.Therefore,in view of the special task of marine plankton research,this paper proposes two no-reference MPI clarity assessment(NR-MPICA)algorithms based on the clarity of image quality.In view of the two aspects mentioned above,the work is summarized as follows:1)NR-SIQA method based on visual saliency and perceptionThis paper proposes a NR-SIQA method based on visual saliency and perception.Firstly,global and local features of left and right views are extracted,and then the perceptual models of visual saliency area and just noticeable difference(JND)of image itself are calculated respectively,and the obtained visual features are weighted by using both of them to reflect the visual differences of different features in the process of visual perception.At the same time,in order to obtain accurate binocular visual perception information,the global structure feature reflecting spatial correlation is extracted from the image synthesized from left and right views,and then the quality evaluation score of stereoscopic image based on this feature is learned by using support vector regression(SVR)model.Experiments on the classic SIQA datasets show that the NR-SIQA method based on JND is superior to the latest quality evaluation method.2)NR-MPICA algorithm based on SVR classification.This paper presents a NR-MPICA algorithm based on SVR.Through the analysis of the definition evaluation algorithm for ordinary images,the specific features in MPIs can be extracted manually for definition evaluation and classification of definition and blur.Firstly,global and local features such as discrete orthogonal moments and local contrast statistics are extracted from MPIs.Then,SVR regression model is used for feature training and regression fitting to get the final prediction clarity.According to the predicted definition value,the subjective definition threshold is used to realize the clear and blur classification of MPIs.The algorithm shows certain accuracy in limited datasets.3)NR-MPICA algorithm based on convolutional neural network(CNN).In order to use more abundant high-level features to achieve clarity evaluation,this paper makes full use of the advantages of deep learning network in image feature extraction,and designs multi-layer convolution neural network to extract deep-level features in MPIs.Then,the trained model is used for clear fuzzy classification,and the classification results are compared with the subjective evaluation results.Experimental results show that this method can achieve higher accuracy than manual feature extraction.