文件名称:3
介绍说明--下载内容均来自于网络,请自行研究使用
基于马氏距离度量的局部线性嵌入算法
局部线性嵌入算法(LLE)中常用欧氏距离度量样本间相似度.而 对于图像等高维数据,欧氏距离不能准确体现样本间的相似程度.文中提出基于马氏距离度量的局部线性嵌入算法(MLLE).算法首先从现有样本中学习到一个 马氏度量,然后在LLE算法的近邻选择、现有样本及新样本降维过程中用马氏度量作为相似性度量.将MLLE算法及其它典型的流形学习算法在ORL和 USPS数据库上进行对比实验,结果表明MLLE算法具有良好的识别性能.
-Based on local linear embedding algorithm Mahalanobis distance metric LLE algorithm (LLE) common Euclidean distance measure of similarity between the samples. For high-dimensional image data, Euclidean distance can not accurately reflect the degree of similarity between samples. In this paper, Mahalanobis distance metric based on local linear embedding algorithm (MLLE). Firstly, learning the existing sample to a Markov measure, then LLE neighbor algorithm selection, existing samples and new samples dimensionality reduction process using Markov measure as a similarity measure would MLLE algorithms and other typical manifold learning algorithm on ORL and USPS to compare experimental results show MLLE algorithm has good recognition performance.
局部线性嵌入算法(LLE)中常用欧氏距离度量样本间相似度.而 对于图像等高维数据,欧氏距离不能准确体现样本间的相似程度.文中提出基于马氏距离度量的局部线性嵌入算法(MLLE).算法首先从现有样本中学习到一个 马氏度量,然后在LLE算法的近邻选择、现有样本及新样本降维过程中用马氏度量作为相似性度量.将MLLE算法及其它典型的流形学习算法在ORL和 USPS数据库上进行对比实验,结果表明MLLE算法具有良好的识别性能.
-Based on local linear embedding algorithm Mahalanobis distance metric LLE algorithm (LLE) common Euclidean distance measure of similarity between the samples. For high-dimensional image data, Euclidean distance can not accurately reflect the degree of similarity between samples. In this paper, Mahalanobis distance metric based on local linear embedding algorithm (MLLE). Firstly, learning the existing sample to a Markov measure, then LLE neighbor algorithm selection, existing samples and new samples dimensionality reduction process using Markov measure as a similarity measure would MLLE algorithms and other typical manifold learning algorithm on ORL and USPS to compare experimental results show MLLE algorithm has good recognition performance.
(系统自动生成,下载前可以参看下载内容)
下载文件列表
鍩轰簬椹