Image Quality Assessment Based On Quality-concerned Factors And Structural Distortion

In course of image gathering, transmission, processing, recording and so on, image quality will be affected by concerned techniques. So, the study of image quality assessment is not only important in theoretical sense, but also widely wanted by variety of applications.This dissertation introduces the importance of image information and the definition of image quality assessment, and summarizes and compares some typical image assessment methods at present, and analyzes the theory of wavelet transfonnation and the characteristics of Human Visual System (HVS), and discusses the relations between them, and shows the importantance and necessity of considering HVS in image quality assessment.Considering concerned factors, the dissertation proposes a new method of image quality assessment. According to the theory of wavelet transformation and the characteristic of HVS and the relations between them, the dissertation carries the visual weight on the wavelet transformation coefficient and extracts three indicators that influence image quality, i.e. brightness, clarity and relevance. Then the dissertation integrates the three indicators and gets the image quality. This method fully integrates not only the wavelet transfonnation and the characteristics of HVS, but also the spatial and frequency domain; the brightness indicator is consistent with the expression in spatial domain, the clarity (blur) refers to the proportion of high-frequency components in the frequency domain, and using high-frequency distortion expression, relevance indicator is using the similarity of two frequency components to expression.The dissertation also proposes a method of image quality assessment based on the structure distortion measurement. Considering the theory of image structure distortion, characteristic of HVS and wavelet transformation, the method carries visual weight on the wavelet transformation coefficients and extracts four indicators: the mean frequency comparison, the frequency contrast comparison, frequency relevance comparison and the frequency edge similarity comparison, whose geometric average forms the image quality assessment. This method is different to the preceding method lies in which it didn't considerate the factors in the spatial domain, but process completely in the frequency domain. After wavelet transformation, each component can be viewed as one approximate of the original image. Then, the mean frequency, standard deviation and covariance in each component are extracted, as well as the direction and amplitudes of...