Full – Reference image quality assessment based on positional structural information

Pasha, Reza and Nooshyar, Mehdi and Jamali, Shahram (2016) Full – Reference image quality assessment based on positional structural information. Masters thesis, university of Mohaghegh Ardabili.

Text (ارزیابی کیفیت تصویر به صورت Full – Reference بر اساس اطلاعات ساختاری محلی) Reza Pasha.pdf Download (769kB)

Abstract

Image quality assessment (IQA), has been becoming an important issue in the field of image processing by the increasing growth of digital image applications throughout the last few decades. Since human beings are the ultimate users in most image-processing systems, the most reliable way of assessing the quality of an image is by human subjective evaluation. Nonetheless, due to the subjective IQA methods being expensive and time consuming in real-world applications, it is necessary to design objective IQA algorithms to assess the image quality automatically and consistently with human perception of quality. Most of the proposed IQA algorithms assume that an undistorted original version of under-evaluation image is fully available. These algorithms are called Full-Reference (FR) IQA. In this thesis, a “positional structural information” based approach for FR IQA is proposed. The positional structural information is a set of information about image structures that is obtained from different positions of an image and includes the perceived information, distortion type and color information of that image. Some findings of the probability and estimation theories are used to represent this information. Furthermore, Two new IQA algorithms are proposed based on above-mentioned approach. The former, uses only the perceived information and involves less computational complexity but the latter is designed on the basis of a composed model and takes advantage of all three kinds of structural information to provide more accurate predictions of the image quality. The results of experiments conducted on LIVE standard image database, indicate that two proposed algorithms assess the quality of images with lower error and higher correlation with the subjective evaluations than the known IQA algorithms and can be used as two robust tools in various IQA applications.

Actions (login required)

View Item