二维线性大间距判别分析及其在步态识别中的应用

2D linear maximum margin discriminant analysis and its application to gait recognition

提出一种二维线性大间距判别分析(Two dimensional linear maximum margin discriminant analysis,2DLMMDA)的投影算法。该算法一方面采用了有效且稳定的大间距优化准则,引入了Laplacian矩阵,保持了特征矩阵的流形结构,且优化域为Laplacian类间散度与Laplacian类内散度之差,能克服Fisher准则带来的小样本问题;另一方面,采用了具有监督信息的判别分析,大大地提高了识别率。为了验证所提出的算法对特征提取的有效性,选择最近邻分类器进行特征分类,最后通过在CASIA(B)步态库上实验。实验结果表明,文中提出的算法具有更高的识别率和识别速度。

In this paper, a novel projection algorithm named 2D linear maximum margin discriminant analysis is proposed. The efficient and stationary maximum margin optimization criterion was used in this algorithm, which introduces Laplacian matrix in order to maintain the manifold structure of the feature matrix, and the optimization criterion is the difference of the Laplacian inter-class scatter and Laplacian intra-class scatter. This algorithm can avoid the small sample size （SSS） problem brought by the Fisher criterion. The discriminant analysis is adopted, which has supervision information and greatly improves the recognition accuracy. In order to verify the efficiency of the proposed method for feature extraction, experiment with the nearest neighborhood （NN） classifier on the CASIA（B） database is conducted. The results show that the proposed method gains a higher recognition rate and faster speed.