Study On Objective Quality Assessment Of Digital Images

With the development of science and technology and the popularity of internet, we can obtain more information from the internet. Nowadays, we can immediately share and watch the images and videos online from all over the world. Although imaging and image processing technology develop rapidly with the progress and innovation of science and technology, there has been a difficult problem, which is how to make machines imitate human visual system so that the image is close to the scene we get through our eyes. Image quality directly decides the accuracy and sufficiency of the information we get. Various types of distortion will be brought while image processing. Therefore, studying effective image quality assessment has very important significance.This paper studies on objective quality assessment of digital images and propose three effective methods. The main contributions of this dissertation are as follows.1. A modified full-reference image quality assessment based on discrete orthogonal moment is proposed. Discrete orthogonal moment plays a vital role in image processing and pattern recognition. Low order moment can extract important features from image blocks. The results getting from discrete orthogonal moment may be negative in Wee’s method, so it is reasonable to take the relative moments difference as the quality score of the image instead of the correlation coefficient moments of reference image and test image. The relative moment differences between the two images can be get by calculating the ratio of the absolute moment differences and the moments value of reference image. We use the relative moment differences between the two images to calculate the quality of the image. Extensive experiments demonstrate that the proposed method can evaluate the image quality accurately and maintain high consistency with the subjective ratings.2. A no-reference image quality assessment method based on discrete orthogonal moment is proposed. Discrete orthogonal moment polynomial is symmetrical and the features of moment transform matrix could reflect the image quality, so we can get the image quality score by calculating these features. Moreover, we integrate human visual system and divide the image into flat blocks and uneven blocks, so the final blockiness is obtained by averaging, weighting and totaling the score of each block. Extensive experiments demonstrate that the proposed method can evaluate the blocking artifacts accurately and maintain high consistency with the subjective ratings.3. A no-reference blocking artifact metric based on the characteristics of color shifting is proposed. Coarse quantization is applied to the color component so that color discontinuities are easily produced, which then causes blocking artifacts. In the past, most algorithms evaluate image quality by converting the color image into gray image, so that it is not taking into account of the chromaticity information. This metric evaluate the blockiness in YUV color space. This method takes into account the impact of the color information on the image blockiness. For image color shifting and color missing, we respectively relative colorimetric difference and relative brightness difference between adjacent block boundary to evaluate the blockiness of the image. Extensive experiments demonstrate that the proposed method can evaluate the blocking artifacts accurately and maintain high consistency with the subjective ratings.The paper has 16 charts, 11 tables and 77 references.