摘要
现实场景下拍摄的视频由于存在各种未知失真类型、缺少参考视频,对此类视频的质量评价是一个十分具有挑战性的任务.近年来,研究人员将人类视觉系统的先验知识融合在质量评价任务中.在此基础上,提出一种考虑背景失真的无参考视频质量评价方法.该方法在考虑视频内容的同时,显著增强了对视频背景中信息丢失问题的敏感度,在特征提取阶段充分考虑背景特征的提取;随后,通过引入结合门控机制的通道挖掘技术,高效整合高低维特征,使特征通道更加精准地聚焦于背景失真细节;最终,利用时序建模模块构建特征的时间维度模型,并通过线性回归方法生成视频质量的客观量化评分.使用SROCC(spearman rank order correlation coefficient)、PLCC(pearson linear correlation coefficient)和RMSE(root mean squared error)等评价指标在公开数据集KoNViD-1k、LIVE-Qualcomm和CVD2014开展实验,结果表明该方法不仅与人类主观感知具有高度相关性,且预测误差较小,有效提升了视频质量评估的准确性和可靠性,能够更贴近地模拟人类对视频质量的直观评价.
Due to the existence of various unknown distortion types and the lack of reference videos,it is a very challenging task to evaluate the quality of such videos.In recent years,researchers have integrated the transcendental knowledge of human visual system into the task of quality evaluation.On this basis,a non-reference video quality assessment(VQA)method considering background distortion is proposed.This method not only considers the video content,but also significantly enhances the sensitivity to the information loss in the video background,and fully considers the extraction of background features in the feature extraction stage.Then,by introducing the channel mining technology combined with gating mechanism,the high and low dimensional features are efficiently integrated,so that the feature channels can focus on the details of background distortion more accurately.Finally,the time dimension model of features is constructed by using the time series modeling module,and the objective quantitative score of video quality is generated by linear regression method.The evaluation indexes such as SROC(Spearman rank order correlation coefficient),PLCC(Pearson linear correlation coefficient)and RMSE(root mean squared error)are used to carried out experiments on KoNViD-1k,LIVE-Qualcomm and CVD2014 datasets.The results show that this method not only has a high correlation with human subjective perception,but also has a small prediction error,which effectively improves the accuracy and reliability of VQA and can more closely simulate human visual evaluation of video quality.
作者
朱文佳
张婷
程茹秋
余烨
Zhu Wenjia;Zhang Ting;Cheng Ruqiu;Yu Ye(Anhui Baichenghuitong Technology Co.,Ltd.,Hefei 230001,China;School of Computer and Information,Hefei University of Technology,Hefei 230000,China)
出处
《南京师大学报(自然科学版)》
北大核心
2025年第3期102-111,共10页
Journal of Nanjing Normal University(Natural Science Edition)
基金
安徽省自然科学基金资助项目(2308085MF216)
国家自然科学基金资助项目(62372153).
关键词
视频质量评价
无参考
背景失真
通道挖掘机制
时序建模
video quality assessment(VQA)
no-reference
background distortion
channel mining mechanism
time series modeling
