Research On Image Quality Assessment Based On The Entropy Of The Differential Image

This is a highly developed information world. The relationship between people and information is very closely which never reached before. It means that the acqusition or transmission of information is not only rely on traditional methods like newspapers, telephone, etc., but also on We Chat, QQ and Taobao which are more convenience and fast. But these apps are based on computers, mobile phones and other digital devices, all of these display information in the form of pictures. In summary, the picture has become the main way to spread information in now days. However, in the process of sending or acquiring an image, there are all kinds of distortion, which seriously affected people's acquiring information from the picture. Image quality assessment(IQA) provides many ways to solve the problem.IQA obtains a metric representing the quality of an image by the knowledge of digital image processing and mathematic. IQA makes the optimization of image processing system and the choosing of high quality images from a large number of pictures possible. So IQA has great significance in real life.This thesis puts forward a new idea through the research on the relationship between entropy and image and applies the idea to the full reference and no reference IQA. The main contents of this thesis are as follows:Firstly, this thesis introduces the background, meaning and development history of IQA. IQA is made up of subjective quality evaluation method and objective quality evaluation method and we introduce their basic concepts, their applications, the performance index of objective quality evaluation method and the image libraries. As the emphasis, objective quality evaluation is divided into full reference(FR), reduced reference(RR), no reference(NR) and it introduces some good algorithms in each fields.Secondly, the concept and some properties of the entropy are introduced. The thesis proposes the idea that the image quality can be measured by the entropy of differential image between the distorted image and its reference image by analyzing the relationship among entropy, image and human vision. It points out the meaning and feasibility in image processing.Thirdly, a new FR IQA method based on the entropy of the differential image is proposed. At this part, by analyzing the relationship between the high frequency information of an image and human visual system(HVS), the thesis points out that the algorithm can be optimized by high-pass filter and the result of the experiment proves its good performance.At last, a new NR IQA method based on the entropy of the differential image is proposed and the algorithm is mainly used to evaluate the quality of blurred image and noise image. In this part, the possibility of constructing a reference image for the distorted image by low-pass filter is discussed. Based on the possibility and the relationship between the distorted image and its reference image, the possibility of taking the advantage of the entropy of their differential image to evaluate the quality of the distorted image is analyzed. In order to optimize the algorithm, we take use of the sensitivity of HVS to the image luminance, image contrast and the spatial frequency of the image to make the result of the algorithm more precise. Then, the consistency between the result of the algorithm and human perception proves the feasibility of the algorithm.