基于多特征的视频场景分类*

针对视频检测效率低下问题，提出了一种基于多特征融合及特征阈值的视频场景分类方法—阈值判定分类法。首先，提取场景视频的平均关键帧。然后，根据其结构化特征以及不同空间结构对场景识别的贡献度对平均关键帧进行划分与重组，得到感兴趣区域及次感兴趣区域；接着，分别提取这两个区域的场景特征，并利用多特征融合技术分别得到两者的综合特征。最后，根据综合特征并利用特征阈值，进行场景动态分类。实验结果表明，该方法充分利用了视频的结构化特征，实验准确率达到80%，在一定程度上证明了该分类方法的有效性。

Video scene classification based on multi-features

In order to solve the problem of low efficiency of video detection, based on multi-features fusion and threshold of the features, this paper proposed a new method about video scene classification, called threshold judgment classification. Firstly, it extracted the average key frame. Then, according to its structural features and the contribution of different spatial structures to scene recognition, it divided and recombined the average key frame, and then obtained the regions of interest and the sub regions of interest; then, it extracted the scene features of two regions respectively, and using multi-features fusion technology to obtain their comprehensive features. Finally, according to the comprehensive features and the feature threshold, it classified the scene dynamically. The experimental results show that the method makes full use of structural features of video, and the recognition rate is up to 80%. To some extent, this method is very effective.