基于判别信息的近邻保持嵌入降维方法

针对传统近邻保持嵌入算法(NPE)侧重保持样本的局部结构,而没有考虑样本类别信息的不足,提出判别局部近邻保持嵌入算法DLNPE.该算法利用样本点的局部结构构造新定义下的类内类间散布矩阵,并以此作为判别信息引入目标函数.在6个真实数据上进行实验,证明了所提算法的有效性.

Dimensionality reduction of neighborhood preserving embedding based on discrimination information

Because traditional neighborhood preserving embedding (NPE) focuses on keeping a sample of local structure without taking category information into account, the discriminant local neighborhood preserving embedding algorithm is proposed. The algorithm use the local structure of samples points to construct the new definition within-class and between-class scatter matrix, which are introduced into the objective function as the discrimination information. DLNPE has good discrimination performance and extracts the features distinguishing force. The samples in the same class preserve inherent neighborhood structural and the samples in different classes are separated from each other after embedding to low-dimensional space. The experimental results on six real data show that the algorithm is effective.