Image Quality Assessment Method Research Based On Sparse Representation

In the process of human beings perceive the world and communicate with each other, visual information plays a significant role. In recent years, thanks to the rapid development of computer technology and Internet technology, digital image is applied in many industries, more and more people begin to shoot, transmit and use it. So, how to assess the image quality efficiently and accurately has become a focus problem. Existing image quality assessment method can be divided into subjective assessment and object assessment. Although subject assessment can get a result accurate and reliable, it's time-consuming and inconvenient, hard to applicant practically. Compared with the subject image assessment, object image assessment has the value for expansion and application. But as a result of image distortion factors such as the cause, type and performance are not identical, object image assessment research is still a significant an challenging job. Combining the sparse representation theory, this dissertation research the object image assessment deeply, the major contents are as follows in general :First,we propose a blind image quality assessment method based on natural scene statistics in time and frequency domain. It utilizes the joint statistics information of gradient magnitude(GM)map and Laplacian of Gaussian(LOG) response and the correlation in the wavelet subbands to compose a feature dictionary, then represent the feature of a test image via sparse coding, finally estimate the differential mean opinion scores(DMOS) of dictionary images and sparse coefficient to get the predicted quality scores.The experimental results show that, the proposed method performs not only fast and accurately but also possess better generalization ability and stability.Second, we proposed a no reference stereoscopic image quality assessment method based on local binary pattern(LBP). Texture feature have effect on image quality, LBP distribution can describe the texture properly. Firstly, estimate the LBP distribution of flat image as the 2-D feature, then calculate the LBP distribution of absolute disparity image, part of the distribution is selected as the stereoscopic feature. Finally, sparse representation is applied to simulate the process of perceive object information, represent the feature of a test image via sparse dictionary, then get the image quality. Experimental results demonstrate the proposed algorithm can assess the stereoscopic image quality effectively.Third, we proposed a full reference image quality assessment method based on human visual system feature. It use wavelet transform to simulate the multi-channel property, and sparse representation to restore the sparsity of mammal V1 receptive field. Sparse the distortion image subband blocks with the dictionary which is composed by reference image subband blocks. Predict the image quality with three self-defined index that describe the sparse representation process. The result on each experiments is stable and accurate, and the method can obtain high correlations with the subjective quality evaluations.