基于时域梯度相似度的视频质量评价模型

计算机工程与科学

基于时域梯度相似度的视频质量评价模型

邱亮1，夏慧明1，储久良2

(1.南京师范大学泰州学院，江苏泰州 225300;2.南京理工大学泰州科技学院，江苏泰州 225300)

收稿日期:2016-04-29 修回日期:2016-06-07 出版日期:2018-04-25 发布日期:2018-04-25
基金资助:
国家自然科学基金(61101226)；江苏省泰州市科技发展计划项目(2014039)

A video quality assessment model based

on gradient similarity of temporal domain

QIU Liang1，XIA Huiming1，CHU Jiuliang2

(1.Taizhou College,Nanjiang Normal University,Taizhou 225300;

2.Taizhou Institute of Science & Technology,Nanjing University of Science and Technology,Taizhou 225300,China)

Received:2016-04-29 Revised:2016-06-07 Online:2018-04-25 Published:2018-04-25

摘要/Abstract

摘要：

频质量评价有助于多媒体网络系统优化和视频编解码算法改进，近年来已成为图像质量评价领域的热门研究方向。在图像质量评价模型(FSIM)的基础上，结合视频局部多帧之间的时域相关性，通过采用新型的三维梯度算子计算原始视频序列与失真视频序列间的梯度相似度矩阵，提出了一种基于时域梯度相似度的视频质量评价模型(TGSMFSIM)。在LIVE视频数据上的测试结果表明，所提模型与视频主观评价有较好的一致性，SROCC与PLCC指标优于VSSIM和VQM两种广泛使用的视频质量评价算法。

关键词: 视频质量评价, 时域相关性, 三维梯度算子, 梯度相似度

Abstract:

Video quality estimation algorithms can be used in multimedia network system optimization and the improvement of video encoding and decoding algorithms, thus they have become a popular orientation of image quality estimation in the past few years. Based on the feature similarity index for image quality assessment (FSIM), using the temporal correlation among video local multiframes, and adopting a new 3D gradient operator to calculate the gradient similarity matrix between the original video sequence and the distorted video sequence, we present a new video quality assessment model based on gradient similarity of temporal domain (TGSMFSIM). The test results on LIVE video data show that the proposed model conforms well to subjective video assessment, and the indexes of SROCC and PLCC are better than those of the VSSIM and VQM, two widely used video assessment algorithms.

Key words: video quality assessment, temporal domain correlation, threedimensional gradient operator, gradient similarity

邱亮1，夏慧明1，储久良2. 基于时域梯度相似度的视频质量评价模型[J]. 计算机工程与科学.

QIU Liang1，XIA Huiming1，CHU Jiuliang2.

A video quality assessment model based

on gradient similarity of temporal domain

[J]. Computer Engineering & Science.