非主观值训练的盲视频质量评价算法
Blind Video Quality Assessment Strategy Without Subjective Scores Training

针对现有基于机器学习的无参考视频质量评价方法中需要利用大量主观评价分值进行训练，导致复杂度高的问题，提出一种非主观值训练的盲视频质量评价算法．首先，利用高斯差分滤波器提取视频结构特征矢量，通过计算与质量感知中心的距离，来评估视频空域感知质量；然后，利用聚类算法获取对运动矢量进行分类的阈值，进而得到运动感知因子；最后，结合视频空域感知质量和运动加权因子得到视频客观质量．实验结果表明：该算法在LIVE video quality 数据库中对视频质量预测的单调性和精确性分别达到了0.817,7和0.828,5，优于对比的其他盲视频质量评价算法；同时，该算法计算复杂度低，易于实现.
Existing no-reference video quality assessment methods based on machine learning needed to use a lot of subjective scores for training which leads to high complexity，thus a blind video quality assessment strategy without subjective scores training was proposed in this paper. Firstly，video structure characteristic vector extracted by the difference of Gaussian (DoG)filter，and the video space perceived quality was estimated by calculating the distance between the structure vector and the quality-aware center. Secondly，the classification threshold of motion vector was obtained by clustering algorithm，and the motion perception factor was acquired. Finally，the video objective quality the video space perception quality and the motion perception factor. The proposed algorithm was tested on the LIVE video quality database. Experimental results show that the proposed algorithm，which possesses a monotonicity and a prediction accuracy of up to 0.817,7 and 0.828,5，respectively，is better than other existing blind video quality assessment methods，and that it is easy to implement because of low computational complexity