Abstract: Content recognition is one main purpose of video and image analysis，however，due to the complexity of video content，it is hard to realize automatic analysis and identification，and there are no universal method and theories of identification of video at present.3D anisotropic diffusion equations can achieve video signals analysis in the multi-scale to facilitate the description and extraction of characteristics.It can solve the video object recognition，activity recognition and behavior recognition etc by the multi-scale analysis and SVM（Support Vector Machines） classification system.The experiments show that the new method can identify video behavior even in the circumstances where sample characteristics remain unknown，which prove that the method can be used to identify any of the known samples through machine learning.

Key words: video, behavior identification, three-dimensional diffusion model, Support Vector Machines（SVM）

摘要： 内容识别是视频和图像分析的主要目的之一，由于视频内容的复杂性，视频内容自动分析和识别并不容易实现，目前没有通用的视频行为识别方法和理论。三维各向异性扩散方程可以实现视频信号的多尺度分析，方便行为特征的描述和提取。通过多尺度分析结合支持向量机分类，可以系统地解决视频对象识别、运动识别和行为识别等一系列问题。实验表明，即使在样本行为特征未知的情况下，通过机器学习，也可以实现对视频行为的识别，说明这种行为识别方法可以用于任意已知样本的行为识别问题。

关键词: 视频, 行为识别, 三维扩散模型, 支持向量机