基于公共向量的模糊邻域保持嵌入算法

摘要

邻域保持嵌入( NPE)算法直接使用K近邻重构样本，由于未区分同类近邻与异类近邻的重要性导致其识别效果不佳，因此提出一种基于公共向量( CV)的模糊邻域保持嵌入算法。首先根据样本K近邻的类别信息求出每个样本对每个类别的隶属度，然后使用公共向量和隶属度重构每个样本，并最小化原始样本与重构样本的残差，最后将该问题转化为求解相应的广义特征值问题以获得最终的投影变换矩阵。该算法尽可能减少投影后同类样本的差异性，较好地分离异类样本。在ORL、Yale、AR和PIEC29这4个人脸数据库上的相关实验验证了算法的有效性。%Neighborhood preserving embedding directly reconstructs the sample by its K-nearest neighbors. However, it does not distinguish the importance between intra-class neighbors and inter-class neighbors, which leads to poor recognition performance. In this paper, a common vector-based fuzzy neighborhood preserving embedding ( FNPE/CV ) algorithm is proposed. Firstly, the degree of membership of every sample for each class is obtained based on the class labels of its K-nearest neighbors. Then, every sample is reconstructed by the common vector and its membership grade for every class. Finally, the problem of minimizing the residual between original sample and its reconstruction sample is converted to solve the generalized eigenvalue problem to obtain the final projection transformation matrix. After the projecting, FNPE/CV minimizes the difference among intra-class samples and separates inter-class samples as far as possible. The experiments on ORL, Yale, AR and PIEC29 face databases demonstrate the effectiveness of the proposed algorithm.