基于自适应密度聚类非线性流形学习降维方法研究与实现

Research and Implementation of Dimension Reduction Method for Nonlinear Manifold Learning Based on Adaptive Density Clustering

针对传统降维方法降维后对流形产生扭曲导致流形展开后的结构发生"畸形"、正确率较低、可信度较差等不足.本文提出了一种基于自适应密度聚类的非线性流形学习降维方法,用分段线性模型来近似流形.利用MATLAB设计并实现了算法,通过实验证明在人造和现实图像数据集上,本流形学习降维算法与现有最先进的流形学习算法相比,产生良好的降维效果.

Aiming at the shortcomings of the traditional dimension reduction method that the manifold is distorted by those method af- ter the dimension reduction causes the ＂deformity＂ of the manifold after the expansion of the manifold. Otherwise, the correct rate and the credibility of traditional dimension reduction method is low. In this paper, a nonlinear manifold learning dimension reduction meth- od based on adaptive density clustering is proposed, and the piecewisc linear model is used to approximate the manifold. The algorithm is designed and implemented by MATLAB. It is proved by experiments that in the artificial and realistic image datasets, the manifold learning dimension reduction algorithm has a better dimension reduction result compared with the existing advanced manifold learning algorithm.