基于单帧图像质量加权的视频质量评价模型  

Video quality assessing model based on single image quality with different weights

作　　者：常青[1] 佟雨兵[1] 张其善[1] 吴今培[2] 

机构地区：[1]北京航空航天大学电子信息工程学院,北京100083 [2]五邑大学智能技术与系统研究所,江门529020

出　　处：《北京航空航天大学学报》2007年第3期311-314,共4页Journal of Beijing University of Aeronautics and Astronautics

基　　金：国家自然科学基金资助项目(60372018);航空科学基金资助项目(04F51068)

摘　　要：利用人眼视觉特性与视频序列时空相关特性,提出了基于视频序列内单帧图像质量加权的视频质量评价模型.其中,单帧图像质量利用峰值信噪比和结构相似性度量作为图像质量的描述参数,采用神经网络(NN,Neural Network)与支持向量机(SVM,Support Vector Machines)建立图像质量评价模型;视频序列质量由序列内单帧图像质量加权衡量,加权因子描述了视频序列内运动及场景变化的剧烈程度.仿真实验结果表明,该模型的输出能有效地反映图像的主观质量.模型预测出的单帧图像质量和视频序列质量的单调性相比PSNR分别提高7.42%和10.47%,均方根误差相比则提高了36.06%和10.48%.Concerning HVS （human visual system） physiological characteristic with the temporal-spatio correlations of video sequence, a video quality assessing model was proposed based on single image quality with different weights. Single image quality was assessed by using NN （neural network ） and SVM （support vector machine） with PSNR （peak signal to noise ratio） and SSIM （structure similarity） as two indexes describing image quality. Video quality was assessed by using the quality of each frames in the video sequence with different weights. Those weights described motion and scene changes in the video. The monotonicity of the method for images ＆ video is 7.42% and 10.47% higher than that of PSNR and RMSE （root mean square error） is 35.90% and 10.48% higher than that of PSNR at least. The results from simulation experiments show the model is valid.

关 键 词：支持向量机 神经网络 视频质量 场景变化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]