基于视敏度的Q-STAR模型参数预测  

The parameters prediction of Q-STAR model based on video acuity

作　　者：杜麟[1] 田畅[1] 吴泽民[1] 张兆丰[1] 胡磊[1] 张磊[1] 

机构地区：[1]中国人民解放军理工大学通信工程学院,江苏南京210007

出　　处：《微型机与应用》2017年第1期40-43,共4页Microcomputer & Its Applications

基　　金：国家自然科学基金(61501509)

摘　　要：Q-STAR模型是现有的客观视频质量评价中最接近主观得分的模型,但是在解码端难以从丢包解码后的YUV视频序列中提取与编码端相一致的运动矢量信息,从而影响模型参数的预测。针对该问题,提出了基于视敏度信息的模型参数预测方法,从YUV视频序列中提取时间域和空间域视敏度特征,并结合Q-STAR模型中提取的特征进行预测。通过实验发现,使用视敏度信息代替原有的运动矢量信息进行预测,所得模型参数值与Q-STAR基本相符,并且计算更为简单。Q-STAR is now the best model whose objective evaluation is close to the real subjective scores. But in the decoding end, it＇ s diffi- cult to extract the same information of motion vector from YUV sequences with packet loss, which influences the prediction of model parame- ters. In order to solve this problem, a new model parameters prediction method was proposed based on video acuity. Firstly, it extracts video acuity features of time domain and space domain from the sequence, and then predicts the model parameters with the features extracted in the Q-STAR. Experimental results demonstrate the new prediction method is close to the old prediction and makes the calculation easy.

关 键 词：视频传输质量保障 主观感知 参数预测 视敏度信息 

分 类 号：TN919.85[电子电信—通信与信息系统]