Face Recognition Using Artificial Neural Network
No Thumbnail Available
Date
2006-08
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
In recent years, an explosion in research on pattern recognition systems using neural
network methods has been observed. Face Recognition (FR) is a specialized pattern
recognition task for several applications such as security: access to restricted areas,
banking: identity verification and recognition of wanted people at airports.
This thesis will explain what is involved in FR task and outline a complete Face
Recognition System (FRS) based on Artificial Neural Network (ANN). In this work, two
FRS are developed. The first model uses Principal Component Analysis (PCA) for
feature extraction from the face images and ANN for the classification purpose. In the
second model, combination of Gabor Filter (GF) and PCA are used for feature extraction
and ANN for the classification.
In the first approach, the face images are projected into subspace called eigenspace,
consisting of the eigenvectors from the covariance matrix of the face images. The
projection of an image into eigenspace will transform the image into a representation of a
lower dimension which aims to hold the most important features of the face. These
feature vectors are classified into training, validation and testing set. The training and
validation set are used during the training of ANN. The testing set is used to evaluate the
recognition performance of the model.
In the second approach, Gabor feature vectors are derived from a set of downsampled
Gabor wavelet representations of face images, then the dimensionality of the vectors is
reduced by means of Principal Component Analysis (PCA), and finally ANN is used for
classification. The Gabor filtered face images exhibit strong characteristics of spatial
locality, scale, and orientation selectivity. These images can, thus, produce salient local
features that are most suitable for FR.
Experimentation is carried out on FRS by using Olivetti Research Laboratory (ORL)
datasets, the images of which vary in illumination, expression, pose and scale. The result
shows the feasibility of the methodology followed in this thesis work. Model 1 achieves a
recognition rate of 76.6% whereas model 2 achieves 88.3% of correct classification and
performed very efficiently when subjected to new unseen images with a false rejection
rate of 0% during testing. The high recognition rate of model 2 shows the efficiency of
GF in feature extraction.
Key words—Face recognition, biometrics, artificial neural network, Gabor filter and
principal component analysis.
Description
Keywords
Face Recognition, Biometrics, Artificial Neural Network, Gabor Filter And Principal Component Analysis