A hybrid approach for recognizing faces

Abstract EN

Most face recognition systems perform poorly with sharp poses.

Sharp poses are images of the face taken after physical rotation with large angles.

In sharp poses, the amount of skin, features, and objects usually vary from one pose to another.

This paper introduces a hybrid approach for recognizing faces from sharp poses.

The approach utilizes two methodologies based on the Principle Component Analysis (PCA) algorithm.

The PCA algorithm learns a set of common features from the training images.

The first methodology utilizes the wavelet transformation to reduce the dimensionality of the images and preserve the important information of each image before applying the PCA algorithm.

The second methodology applies the PCA algorithm which represents each training image as weighted vectors in terms of the important set of Eigen faces.

A linear discriminate analysis (LDA) algorithm is used to maximize the distance among weighted vectors from different classes and minimize the distance among weighted vector from the same class.

For each methodology, twenty different experiments were performed.

The obtained results show that combining PCA and LDA did not improve the accuracy, while combining the wavelet transformation and PCA were faster and more accurate.