摘要： 在局部线性嵌入算法中，标签价值没有得到充分体现。针对该问题，提出一种基于核的半监督局部线性嵌入方法。考虑到欧氏距离容易破坏流形结构，将原始数据映射到高维核空间，利用高维空间中的核距离代替欧氏距离，采用半监督标签信息调整距离矩阵，通过调整后的距离矩阵对数据结构进行线性重建，从而提高算法的降维性能。在标准数据集、人脸库、字符库等数据上进行实验，结果表明，与传统局部线性嵌入算法相比，该方法的辨识率提高了2%

关键词: 流形学习, 半监督学习, 局部线性嵌入, 人脸识别, 字符识别

Abstract: In order to solve the defects that Locally Linear Embedding(LLE) can not make full use of lable information in unsupervised machine learning, this paper proposes a kind of semi-supervised kernel-based local linear embedding algorithm. Taking into account measure of Euclidean distance is easy to destroy the manifold, this algorithm maps the raw data into a high dimensional kernel space, uses the distances in high-dimensional space instead of euclidean distance, and introduces the thought of semi-supervised learning to adjust the distance with label information, which is used for linear reconstruction and dimension reduction, and enhances the ability on dimension reduction. In the standard data sets, face and character database, experimental results show the recognition rate of novel method is 2% higher than the traditional local linear embedding algorithms.

Key words: manifold learning, semi-supervised learning, Locally Linear Embedding(LLE), face recognition, character recognition

中图分类号: 

• N945