基于稀疏张量判别分析的人体行为识别

在模式识别中，如何在提取关键特征的同时对样本进行降维与识别是研究的热点之一。在局部Fisher判别分析（LFDA）的基础上，结合张量表示和稀疏分析，本文提出一种基于稀疏张量的特征提取方法：稀疏张量局部Fisher判别分析（STLFDA）。该方法把张量局部Fisher判别分析（TLFDA）算法中特征分解问题转化为线性回归问题，并用弹性网络解决线性回归中的特征选择问题，既满足了张量局部Fisher判别分析的目标，又保证了得到的投影矩阵的稀疏性。通过在Weizmann人体行为数据库上的实验，表明了稀疏张量局部Fisher判别分析（STLFDA）算法的有效性。

Human Behavior Recognition Based on Sparse Tensor Discriminant Analysis

In pattern recognition, how to reduce the dimension and identify the samples while extracting the key features is one of research hotspots. Based on local Fisher discriminant analysis (LFDA), this paper proposes a feature extraction method combining tensor representation with sparse analysis: Sparse Tensor Local Fishers Discriminant Analysis (STLFDA). This method transforms the feature decomposition problem in tensor local Fisher discriminant analysis (TLFDA) algorithm into linear regression problem, and solves the feature selection problem in linear regression with elastic network. It not only satisfies the goal of the Tensor Local Fisher Discriminant Analysis, but also guarantees the sparsity of the projection matrix. The validity of STLFDA algorithm is demonstrated by experiments on the Weizmann human behavior database.