No-reference Image Quality Assessment Method Based On Sparse Representation And Its Application

It is easy for images to produce distortion in the process of collection, analysis, transmission, processing and reconstruction. And it's particularly important for image processing system to identify and quantify the image quality level. FR(Full-Reference) image quality evaluation method is completely dependent on the original image information while RR(Reduced-Reference) image quality evaluation method needs to extract some features from the origin image. But it is difficult to get perfect and distortionfree images in practical application. So the research of NR(No-Reference) image quality evaluation algorithm becomes a hot and difficult subject.As a new representation signal, sparse representation is affected by the visual perception of the cerebral cortex and can be used for sparse coding, so that the high dimensional data is converted to low dimensional data, and finally achieve the purpose of simplifying the signal processing. The most important idea of this model is that the test images can be represented by similar samples subspace of the training samples, and the coefficient will be concentrated sparsely in the training samples.To develop a no-reference image quality assessment system which is highly efficient, accurate and compliance with the HVS(Human Visual System) is still very difficult so far. This thesis proposed a NR quality evaluation algorithm based on sparse representation and its application on the mobile phone screen display effect. Main work as follows.According to the visual characteristics, the thesis introduces the theory of sparse representation and designs a NR image quality evaluation model based on sparse representation. The model extracts the statistical features in spatial domain, constructs features dictionary, and quantifies the quality of the image by the sparse coefficient. Experimental results show that the subjective quality scores and objective quality scores obtained by the corresponding image quality evaluation model can keep consistency well.In order to study the application of the NR quality assessment model in the mobile phone screen display effect, the thesis proposes a content adaptive backlight control technology for mobile devices. The selected images are filtered by the NR image quality assessment model. Next, according to the attention mechanism of human vision, it extracts the ROI(Region of Interest) of the image, then analyzes content information of the ROI and extracts the representative image features. According to the human visual experiment to get the most suitable screen brightness, and the regression model is built between the image characteristics and the screen brightness. And the screen backlight brightness of the test image can be obtained by the regression model. Experimental results show that the method can effectively adjust the backlight of the screen according to the contents information of the screen.