Spatial No-Reference Image Quality Assessment

With the extensive application of digital image and video, the issue of digital image quality assessment becomes more and more important. As a basic problem of image processing, image quality assessment is not only important in theory, but also wanted in wide backgrounds of application. Digital image quality assessment usually can be divided into subjective assessment and objective assessment. Subjective assessment method is tedious, time-consuming and can not be implemented in real-time. Objective assessment methods are classified into three categories: Full-Reference quality assessment (FR-QA), Reduced-Reference quality assessment (RR-QA), and No-Reference quality assessment (NR-QA). Since full or partial information are required in FR-QA and RR-QA, they are limited for practical applications. For this reason NR-QA should be adopted.At first the related researches of image quality assessment have been briefly summarized, especially no-reference methods. Generally speaking, on the one hand, mostly human is the ultimate receiver of image and the results of image quality assessment should be agree with human's feelings, so HVS properties are introduced in many assessment methods. On the other hand, however, on certain occasions, such as machine vision, which still can not be mentioned in the same breath with the human's complex sense, it must be paid more attention to the quantitative models and parameters. Therefore, the dissertation proposes to assess image quality from the two aspects, i.e. human vision-oriented and machine vision-oriented respectively. The former mainly considers from the human perception, and the latter takes the models and parameters into account, which is easy to process for machine, and the corresponding frameworks of the general system are also presented. In order to satisfy the needs of real-time processing, it will be mainly handled from spatial domain. By analyzing the factors that could influence image quality, several typical factors which have significant impact on image quality are extracted: contrast, definition, noise. These quality factors are discussed separately, since on the one hand, performance study of the single factor will facilitate to embed the single-factor into assessment system, and on the other hand, it can help sorting out the effects of interaction between these factors, in order to form a unified synthetical index.For the human vision-oriented applications, human visual system and habits are mainly considered. With a view of blur, based on the smoothing effect of blur along edges or in textured areas and the first-order derivative of the spread neighborhood on...