Research On Image Quality Assessment Based On Visual Perception

Image quality assessment is to analyze the image intelligently and predict the perceived image quality automatically based on designing a mathematical model. It can be classified into full reference, reduced reference and no reference according to the availability of the original image. It becomes one of the key technologies for benchmarking the image compression or processing algorithms and plays an important role in the multimedia application system. Meanwihle, the image is the basic unit of the video, and the video composes of the images which are arranged with time sequence. The research on the image quality assessment is the important basis of the one on the video quality assessment. However, the traditional subjective image quality assessment which depends on the observation of human beings is impractical, slow and expensive for most applications. Researches focusing on developing the new generation of image quality assessment which can reflect the subjective feeling of human beings precisely and can also be applied in the practical multimedia systems become a necessity. Currently, Human Visual System (HVS) is the only way to access the image information. The perception of HVS for the image is selective, and different regions or objects in the image have diverse levels of visual importance. However, the current image quality assessment ignores this diversity of perception mechanism. Therefore, it is of theoretical significance and practical value to take an in-depth research on the improvement of the prediction accuracy of image quality assessment by applying the principle of the visual perception.In chapter 1, the significance of the research work is presented together with a brief summary of the present research status.In chapter 2, a structural similarity quality metric based on the visual perception is proposed, which is a full reference metric. The spatial features are combined to produce the visual perceptual map. The focus of attention is shifted because of the most visual saliency or a serious distortion; it results in the reproduction of the visual perceptual map. The structural similarity is weighted by the above two visual perceptual maps, and the objective quality of the image is acquired.In chapter 3, an image quality assessment based on watermarking in the wavelet domain is proposed, which is a reduced reference metric. The watermark is embedded into the selected frequency sub-bands according to the different distortion sensitivity. Meanwhile, the watermark embedded map and the adaptive adjustment of quantization parameter are applied to decide the position and the strength of the watermark, which ensure the watermark's invisibility effectively. The image objective quality is acquired through the watermark recovery rate.In chapter 4, a perceptual blockiness metric and an Orthogonal Least Squares Radial Basis Function (OLS-RBF) based blurring and ringing metric are proposed; they are both no reference metrics. The blockiness metric calculates the spatial visual features to produce the visual perceptual map. Then, it weighs the blockiness to acquire the image objective quality. The blurring and ringing metric extracts the generalized features of the edge points in the structure-texture region; the objective quality of the JPEG2000 image is evaluated by training an OLS-RBF network model from the generalized features and the subjective scores.In chapter 5, a video structural similarity quality assessment based on the visual perception is proposed, which is a full reference metric. A distortions-weighing spatiotemporal visual attention model is designed. The visual perceptual map which is produced based on this model and the structural similarity map are used for the objective quality of the single video frame. Then, the objective quality of the whole video sequence is calculated by introducing a frame quality contribution function which weighs the qualities of each frame and gives a much heavier weighting factor to the extremely damaged frames to takes the case of burst-of-error into account.The final chapter concludes the new achievements of the whole research and the prospect of the future research.