Non-reference assessment of video quality using statistical methods

DOI: 10.31673/2412-4338.2020.043543

Abstract

The article considers the problem of finding a way to assess the quality of video in the absence of a standard for comparison. In the literature, such methods of assessing image quality are called no-reference (NR) or NR-methods. First of all, the article examines the artifacts of image compression. The relevance of this approach is that the data is compressed when material transmitting over the Internet to save information. This method is based on criteria that characterize the degree of change in the brightness of video frames. By themselves, the criteria allow to conduct a comparative analysis of image quality not in all cases. In this article, to assess the quality it is proposed to use criteria which are based on statistical methods, which reflects the degree of change in brightness in the aggregate. These criteria are completely new in the field of research the quality of both video streaming and images in general. The proposed method takes into account all possible changes in the characteristics of the image with deteriorating quality. During the experiment, the feasibility of using these methods in the problem of ranking the material by the level of compression artifacts was demonstrated. It has been experimentally shown that none of the studied non-reference methods of image quality assessment is universal, and the calculated assessment cannot be converted into a quality scale without taking into account the factors influencing the distortion of image quality. Also, this method forms the final estimate as the arithmetic mean of the estimates of rows and columns of the image. In the case of local distortions, the proposed methods may not give completely true results. To conduct the experiment, the program code was implemented in the MATLAB environment, using the library for computer image processing Image Processing Toolbox.

Keywords: video quality assessment, standard deviation, coefficient of variation, coefficient of oscillation, relative linear deviation, relative rate of quartile variation, compression artifacts.

References
1. Parker J.R. (1996) “Algorithms for image processing and computer vision”.
2. Wesley E. Snyder, Hairong Qi. (2017) “Fundamentals of Computer Vision”. Published by Cambridge University Press, University Printing House, Cambridge CB2 8BS, UK.
3. Grebenyuk V.V.(2020) Comparative analysis of non-reference methods for assessing the quality of video material. Communication №6 (142). P.34-42.
4. Adrian Kaehler, Gary Bradski (2016). “Learning OpenCV 3. Computer Vision in Matlab”. Published by O’Reilly Media, Inc., 1005 Gravenstein Highway North, Sebastopol, CA 95472.
5. Prateek Joshi, David Millan Escriva, Vinicius Godoy (2016). “OpenCV By Example”. Published by Packt Publishing Ltd. Livery Place, 35 Lively Street, Birmingham B3 2PB, UK.
6. Acree, M.C. (1978). “Theories of Statistical Inference in Psychological Research: A Historicocritical Study”. Dissertation. University Microfilms International H790 H7000, Ann Arbor, MI.