基于区域属性特征和多核支持向量机的运动视频分类

Motion video classification based on regional attribute features and multi-core support vector machine

为了提高视频搜索平台中海量运动视频数据的分类精度,提出了一种基于区域属性特征和多核支持向量机的运动视频分类方法。首先,利用区域分块策略对输入视频进行初步处理,并提取颜色特征以便快速获取视频中的关键帧。然后,根据每个区域中的运动属性、纹理属性等因素来提取视频的主要特征。最后,采用权重的方式将局部和全局的核函数进行线性相加,组成多核支持向量机实现视频分类。在6种类型运动视频构成的数据集上进行了分类测试。实验结果表明,相比于其他类似的支持向量机方法,提出方法在查全率、查准率和F1三个分类评估指标上均有一定提高,平均分类F1指标达到92%。

In order to classify massive sports video data quickly and effectively and build a more intelligent video search platform,a sports video classification method based on regional attribute features and multi-core support vector machine is proposed.Firstly,the input video is preliminarily processed by using the regional block strategy,and the color features are extracted to quickly obtain the key frames in the video.Then,the main features of the video are extracted according to factors such as motion attributes and texture attributes in each region.Finally,the local and global kernel functions are linearly added by weight,and a multi-core support vector machine is formed to realize video classification.Experimental results show that,compared with other similar support vector machine methods,the proposed method can improve the recall ratio,the precision ratio and the classification evaluation index of F1,and the average classification index of F1is 92%.