基于混沌特征的运动模式分割和动态纹理分类

Chaotic Features for Motion Pattern Segmentation and Dynamic Texture Classification

采用混沌理论对动态纹理中的像素值序列建模,提取动态纹理中的像素值序列的相关特征量,将视频用特征向量矩阵表示.通过均值漂移(Mean shift)算法对矩阵中的特征向量聚类,实现对视频中的运动模式分割.然后,采用地球移动距离(Earth mover s distance,EMD)度量不同视频的差异,对动态纹理视频分类.本文对多个数据库测试表明:1)分割算法可以分割出视频中不同的运动模式;2)提出的特征向量可以很好地描述动态纹理系统;3)分类算法可以对动态纹理视频分类,且对视频中噪声干扰具有一定的鲁棒性.

In this paper, we propose a novel framework for dynamical texture modeling based on chaos theory. Our method first extracts features from dynamical texutre and concatenate the features to a feature vector. A video is then represented by a feature matrix. The mean shift clustering algorithm is used to cluster the feature vector which achieves segmenting videos into different motion patterns. The earth mover＇s distance （EMD） is employed to compute the feature cluster similarities and classify the dynamic textures. Experimental results indicate that： 1） The segmentation algorithm can cluster different motion patterns in videos; 2） The feature vector proposed in this paper can effectively characterize the dynamical texture; 3） The proposed algorithm can classify dynamical texture accurately. In addition, the algorithm is robust to video noise.