Research On No-reference Authentic Image Quality Assessment

Nowadays, with the rapid development of science and technology,people’s living standards are improving. Meanwhile, the multimedia applications are gradually expanding. At present, the image and the video has already been an import tool to exchange information in daily life and entertainment. Therefore, in order to improve the user experience, the study of real distorted image quality evaluation has great practical significance.In the transmission process, the heterogeneity of the transmission network,dynamic bandwidth, will inevitably lead to image distortions. These distortions will seriously affect the image quality, reduce the user experience, thus restricting the development of image applications.Generally, the development of the image quality evaluation method is based on the single distorted image. So we firstly study the classic and effective image quality algorithm which is based on the single distortion,then improve the algorithm to be applied on the real distorted images. The following is the research results of this paper:1) In the paper, we propose a general non-reference real distorted image quality algorithm based on human visual aesthetic perception. The algorithm firstly extracts the natural scene statistical features of the image in the spatial domain, and then extracts the aesthetic features of the image,such as fuzzy, contrast, dynamic range, color information and so on. Then the support vector machine is used to regress the selected features.Experimental results show that the proposed method has good performance and can achieve better consistency with subjective perception.2) In the paper, we also make a main motion blur library of real distorted images to study the motion blur features, and propose a special model based on the main motion blur image library. The model optimizes the extraction of fuzzy features, extracts the statistical features of natural scenes and human visual aesthetic features. The experimental results show that the model has relatively good consistency with human visual perception.