Quaternion model of video quality assessment weighted by gray-level co-occurrence

Abstract

In this paper, we proposed a new video quality metric model, which describe the video quality feature using quaternion matrix. The four parts of the quaternion are brightness, chrominance, contour, and inter frame residual, respectively. The feature value of each block is computed from quaternion model by singular value decomposition. In addition we extract gray-level co-occurrence matrix (GLCM) of each block, then calculate differential entropy of GLCM as the weight of the block. The algorithm is tested on the video quality expert group (VQEG) Phase I FR-TV test data set.Experiments show that it has good correlation with perceived video quality. Keywords:Video quality assessment; quaternion; Gray-Level Co-occurrence Matrix; video quality expert group (VQEG) IntroductionDigital videos are increasingly using in people's day-to-day lives. There are many applications of videos such as internet videos, digital television, video conferencing, and so on. Methods for evaluating video quality are become more and more important to maintain the quality of internet service, or to evaluate the perfonnance of the compression. The best method to evaluate quality of videos is subject evaluation, but it is complex and much time cost. The objective evaluation methods become a focus of the video evaluation.In this paper, we describe the quality information of videos using quaternion matrix, which include four parts to represent main features of the video. According to human visual theory, contour information and surface infonnation are important in distinguishing objects[ I ] [2]. In the quaternion model, chrominance information also used to describe contour and surface feature in three parts, not only luminance infonnation. inter-trame-difference (TFD) is used to describe temporal domain feature of the videos. In 978-1-4673-4685-6/12/$31.00 ©2012 IEEE 193 addition, we calculate differential entropy of the gray-level co-occurrence matrix [3] [4] as the weight of the block, to evaluate level of surface detail of blocks. Finally, The quaternion is calculated by singular value decomposition to measure the the quality value of the video. Experiments show this evaluation model is superior to the traditional method, such as MSE, PSNR and VSSIM [5] .The rest of the paper is organized as follows. In section 2, we give a brief introduction about Quaternion matrix and singular value decomposition (SVD); in section 3, we describe how to build quaternion matrix for video quality assessment; section 4 describes how to compute differential entropy of the gray-level co-occurrence matrix as the weight of a block; section 5 gives out the steps to calculate feature value of the videos; in section 6, we compare the results of different video quality assessment models tested on the video quality experts group (VQEG) Phase I FR-TV video dataset; finally, Section 7 draws conclusions. Quaternion and singular value decompositionA quaternion matrix Q is made of one real part and three imaginary parts: Q = a + bi + cj + dk (I) Where i, j, k obey the rules ...