基于支持向量回归的无参考MS-SSIM视频质量评价模型  

No-reference MS-SSIM Video Quality Assessment Model Based on Support Vector Regression

作　　者：卓力[1] 张美娜[1] 王贯瑶 李嘉锋[1] ZHUO Li;ZHANG Meina;WANG Guanyao;LI Jiafeng(Beijing Key Laboratory of Computational Intelligence and Intelligent System,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学计算智能与智能系统北京市重点实验室,北京100124

出　　处：《北京工业大学学报》2018年第12期1486-1493,共8页Journal of Beijing University of Technology

基　　金：国家自然科学基金资助项目(61531006)

摘　　要：多尺度结构相似度(multi-scale structural similarity,MS-SSIM)是一种常用的全参考视频质量评价准则,由于评价时需要原始视频作为参考,因此无法用于实时的网络视频质量评价中,故提出一种基于H. 264码流的无参考MS-SSIM视频质量评价模型.该模型从H. 264码流中提取出I帧和P帧的编码模式、运动矢量等参数,然后对这些参数进行统计分析,来表征视频的纹理丰富程度和运动剧烈与复杂程度;结合量化参数等信息构成码流特征参数集,使用支持向量回归(support vector regression,SVR)方法建立码流特征参数和MS-SSIM之间的映射关系模型,用于预测H. 264码流的MS-SSIM视频质量度量.该模型只使用从H. 264码流中提取的编码参数,无须原始的参考视频,也无须对视频进行解码.与现有的无参考码流预测模型相比,该模型可以获得更高的预测精度.Multi-scale structural similarity(MS-SSIM)is a commonly used full-reference quality assessment metric.Since the original video is required for reference,it is not suitable to be applied in real-time assessment of network video quality.In this paper,a no-reference MS-SSIM video quality assessment model based on H.264 bitstream was proposed.First,I-frame and P-frame encoding mode and motion vector parameters from H.264 bitstream were extracted from H.264 bitstream,which then were statistically analyzed to characterize the richness of the texture and the intensity and complexity of the motion of the video.Second,the parameters were combined with quantization parameter to form the bitstream feature parameter set,and support vector regression(SVR)was finally applied to the relationship between the bitstream feature parameters and MS-SSIM to predict the video quality metric of MS-SSIM for H.264 bitstream.The proposed model only uses the parameters extracted from the video bitstream and does not demand to decode the video bitstream completely.Compared with a state-of-the-art no-reference bitstream prediction model,the proposed model can achieve higher prediction accuracy.

关 键 词：H.264 码流参数 无参考 支持向量回归 多尺度结构相似度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]