A Study on Improving the Mutual Subspace Method by Combining Canonical Angles Using Kernel Fisher Discriminant Analysis

Masatsugu ICHINO  Hitoshi SAKANO  Naohisa KOMATSU 

Publication
D - Abstracts of IEICE TRANSACTIONS on Information and Systems (Japanese Edition)  Vol.J93-D  No.8  pp.1340-1352
Publication Date: 2010/08/01
Online ISSN: 1881-0225
Print ISSN: 1880-4535
Type of Manuscript: Special Section PAPER (Special Section on Image Recognition and Understanding)
Category: 
Keyword: 
canonical angle,  mutual subspace method,  kernel fisher discriminant analysis,  kernel mutual subspace method, 

Full Text(in Japanese): PDF(723.5KB)

Summary: 
We propose a recognition method based on combining the canonical angles calculated by Mutual Subspace Method between subspaces by using the kernel fisher discriminant analysis. In the research of Mutual Subspace Method, the method of using the mean value of the similarities corresponding to canonical angles was proposed. It is probable that each canonical angle has each feature. In this research, some canonical angles are considerd as input patterns. And combining the canonical angles is treated as a pattern classification problem. We have investigated the distributions of the similairty and experimentally obtained their nonlinear decision boundary. So, we have introduced a recognition method of combining canonical angles using the kernel fisher discriminant analysis. We experimentally demonstrate the proposed method's effectiveness with simulation results and show that the method achieved high accuracy.