641.txt

This complete matlab for neural network

发信人: GzLi (笑梨), 信区: DataMining
标  题: [转载] A new milestone for face recognition - laplacianf
发信站: 南京大学小百合站 (Wed Aug 21 09:23:49 2002), 站内信件

【 以下文字转载自 AI 讨论区 】
【 原文由 cloud 所发表 】


we just proposed a new algorithm for subspace analysis (Dimension reduction w
ith new algorithm "LLE").  There are several significances of the algorithm L
LE. 

1. This is the first linear dimension reduction algorithm to approximate non-
linear manifold

  In the state of art algorithm for linear dimension reduction, PCA and LDA a
re dominant two algorithms. However, both of them are derived from Euclidean 
space, rather than manifold. But it has been known that the faces are distrib
uted on a nonlinear 
manifold. Therefore, the optimal algorithm for face recognition should be der
ived directly from the perspective of manifold, rather than Euclidean space.

 

2. The basis obtained by PCA are orthogonal. While LLE will produce non-ortho
gonal basis, just like LDA, ICA, etc. Therefore, it is be able to discover mo
re complex topology of face manifold.


3. LLE has much more power to keep the discriminating information. Moreover, 
it can be also used to do discriminating analysis, just like LDA.


We will submit a paper about using LLE algorithm to do face recognition. It's
 title will be "face recognition with lapacianface"
 

--
※ 来源:．南京大学小百合站 bbs.nju.edu.cn．[FROM: 207.46.71.13]
--
※ 转载:．南京大学小百合站 bbs.nju.edu.cn．[FROM: 211.80.38.29]