Modified nearest feature line method in a 3-D face recognition system for a large number of object classes

Lina, Benyamin Kusumo Putro, Hiroshi Murase

Research output: Contribution to journalArticlepeer-review

Abstract

Authors have developed a novel method for achieving higher recognition capability of a 3-D face recognition system based on feature line method, which is called Modified Nearest Feature Line. Combined with our developed view-based Karhunen-Loeve transformation method as a feature extraction subsystem, the Modified Nearest Feature Line method is used as a classifier to build a 3-D face recognition system. As recognition rates are usually decreased by increasing the number of object classes, the authors evaluate and analyze the performance of this Modified Nearest Feature Line method for recognizing 3-D face images and compared with that of the conventional Nearest Feature Line method. In our experiments, each object class consists of images of persons with their viewpoint positions and expressions. Experimental results show that increasing the number of object classes influenced the recognition rates of both systems. However, the decrement slopes of the recognition system using Modified Nearest Feature Line method were lower than that of using Nearest Feature Line method. It is also shown that at every same number of persons to be recognized, our Modified Nearest Feature Line method always gave a high recognition rate than the original Nearest Feature Line method, with up to 20% in recognition rate difference.

Original languageEnglish
Pages (from-to)338-344
Number of pages7
JournalWSEAS Transactions on Circuits and Systems
Volume4
Issue number4
Publication statusPublished - 1 Apr 2005

Keywords

  • 3-D face recognition system
  • Eigenspace representation
  • Karhunen-Loeve transformation
  • Modified Nearest Feature Line method
  • Nearest Feature Line method

Fingerprint Dive into the research topics of 'Modified nearest feature line method in a 3-D face recognition system for a large number of object classes'. Together they form a unique fingerprint.

Cite this