Abstract: In order to improve the correctness of dimensionality reduction algorithms based on Locally Linear Embedding(LLE) caused by data density change, a novel approach based on density is proposed in this paper. It adapts cam distribute to find the data’s nearest neighbor, meanwhile, adds the data’s density information during the low dimensional local reconstruction. The proposed algorithm is used to reduce the dimensionality of input feature, and the reduced feature is classified by simple classifier. Experimental result indicates that the method can effectively improve the recognition rate of handwritten digits and can dig the manifold embedded in the high dimensional space.

Key words: manifold learning, dimensionality reduction, density information, handwriting recognition, cam distribution

摘要： 针对LLE算法在数据密度变化较大时很难降维的问题，提出一种基于密度刻画的降维算法。采用cam分布寻找数据点的近邻，并在低维局部重建时对数据点加入密度信息。对手写体数字图像进行字符特征的降维，再对降维后的特征进行分类识别。实验结果表明，该方法能区分字符，具有较好的识别率，能够发现高维空间的低维嵌入流形。

关键词: 流形学习, 降维, 密度信息, 手写体识别, cam分布

CLC Number: 

• TP311.13