Improvement Of Face Recognition Using Principal Component Analysis And Moment Invariant

Face recognition attracts many researchers and has made significant progress in recent years. Face recognition is a type of biometric just like fingerprint and iris scans. This technology plays an important role in real-world applications, such as commercial and law enforcement applications, from he...

Full description

Saved in:
Bibliographic Details
Main Author: Thomas, Annie
Format: Thesis
Language:English
English
Published: 2007
Subjects:
Online Access:http://psasir.upm.edu.my/id/eprint/5270/1/FK_2007_58.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Face recognition attracts many researchers and has made significant progress in recent years. Face recognition is a type of biometric just like fingerprint and iris scans. This technology plays an important role in real-world applications, such as commercial and law enforcement applications, from here comes the importance of tackling this kind of research. In this research, we have proposed a method that integrates Principal Component Analysis (PCA) and Moment Invariant with face colour in gray scale to recognize face images of various pose. The PCA method is used to analyze the face image because it is optimal with any similar face image analysis and it has been employed to extract the global information. The vectors of a face in the database that are matched with the one of face image will be recognized the owner. If the vector is not matched, the original face image will be reconsidered with moment invariant and face colour in gray scale extraction. Then, the face will be rematched.In this way, the unrecognized faces will be reconsidered again and some will be recognized accurately to increase the number of recognized faces and improve the recognition accuracy as well. We have applied our method on Olivetti Research Laboratory (ORL) database which is issued by AT&T. The database contains 40 different faces images with 10 each face. Our experiment is done by using the holdout to measure the recognition accuracy, as we divided about 2/3 of the data 280 faces for training, and about 1/3 which is 120 faces for testing. The results showed a recognition accuracy of 94% for applying PCA, and 96% after reconsidering the unrecognized patterns by dealing with pose-varied faces and face colour extraction. Our proposed method has improved the recognition accuracy with the additional features extracted (PCA + face colour in gray scale) with the consideration of the total time process.