【摘要】 邻域保持嵌入通常被广泛用于发现高维数据的固有内在维数。为了充分利用样本的类别信息,构建了一个具有判别信息的邻接矩阵,其可以使同类样本点更加紧凑而异类样本点更加疏远。在此基础上,提出了基于类别信息的邻域保持嵌入算法。基于类别信息的邻域保持嵌入算法在不破坏原始高维数据局部几何结构的同时,可以使处于不同子流形上的样本点尽量分开。在UCI数据集和ORL人脸数据集上的实验结果表明,基于类别信息的邻域保持嵌入算法具有较高的识别率。更多还原

【Abstract】 Neighborhood preserving embedding(NPE)is widely used for finding the intrinsic dimensionality of the data with high dimension.In order to make full use of the classification information of samples to get optimal features,we constructed an adjacent matrix which can separate different sub-manifolds as far as possible without destroying local geometry structure of the original data.By introducing the adjacent matrix,this paper proposed label information-based neighborhood preserving embedding(LINPE).Experiments on UCI data and ORL face databases were performed to test and evaluate LINPE.Experimental results demonstrate the effectiveness of LINPE.更多还原