自适应近邻的局部线性嵌入算法

在局部线性嵌入算法(LLE)中寻找最优近邻数常用试凑法进行搜索,需要大量的时间才能得到最优结果.为此提出基于自适应近邻的局部线性嵌入算法( ANLLE),算法首先给出一个相似性度量函数,然后据此为各个样本设定阈值,根据每个样本周围数据分布情况,为每个样本自动设置不同近邻数,最后在各个样本近邻数不相同情况下进行数据降维及待测样本的分类.在人脸数据库及手写数字数据库上的对比实验表明,ANLLE算法识别性能高于标准LLE算法及邻域线性嵌入算法(NLE).

Adaptive neighborhoods based locally linear embedding algorithm

Finding out the optimal neighbors in Locally Linear Embedding algorithm is still an open problem. Trial and error, a commonly used method, needs much time to get the optimal result. We propose Adaptive Neighborhoods based Locally Linear Embedding algorithm (ANLLE), which firstly provides a new similarity measure function. Secondly, the algorithm sets a threshold for each sample, and then sets various neighbors for each sample according to different distribution around it. Finally, ANLLE reduces the dimensionalities of samples and classifies the out-of-samples in the case of different neighbors for each sample. Comparison of ANLLE, Neighborhood Linear Embedding algorithm(NLE) and standard LLE Algorithm in databases proves that ANLLE is more effective than standard LLE algorithm and NLE algorithm.