3.2.3 Feature Extraction

2.3.3 Feature Extraction
Feature extraction deals with the extraction of the distinctive features out of face images so that those face images can be differentiated among each other. There are several algorithms available for extracting features out of a face image. The most common is the use of mathematical formulas that generate a mathematical representation of a face image that is termed as a template, these templates are “the refined, processed and stored representation of the distinguishing characteristics of a particular individual” (Ted & Neil, 2009).
Many studies has been made to help identify which among the available feature extraction algorithms to use, to improve and how to describe them. Some of the common techniques used for
The whole idea behind Principal Component Analysis (PCA) is that it reduces the dimension of a data set of high dimension and complexity to a data set of less complexity and dimension Principal Component Analysis (PCA) is an efficient method of locating patterns in data set, and transforming the data in such a way that will show the difference and variance within the data set (Shlens, 2003).
The main advantages of principal component analysis in face recognition is that: it eliminates the redundancy of data given the orthogonal components, it reduces the complexity in images grouping, it is very efficient in the representation of small databases and it reduce the noise in an image since the maximum variation basis is chosen and so the small variations in the background are ignored.
The disadvantages of principal component analysis in face recognition is that: the covariance matrix is difficult to be evaluated in an accurate manner and the Principal Component Analysis (PCA) could not capture even the simplest invariance unless the training data explicitly provides this information.
The algorithm follows the following
⅄V=GV 2.6
To implement the principle component transformation algorithm following 7 steps are used. Get or create a database of images for training a classifier Calculate the mean image of the database Subtract the mean from the each image in the database to form an adjusted database Calculate the covariance matrix of the adjusted database Calculate the eigenvectors and the corresponding eigenvalues of the covariance matrix Choose the components (new features) to form a feature space. Using Karuhnen-Loeve Transform to transform the database to the face space. (training the classifier)
Step 1: The first step deals with obtaining or creating a database of images, each image in the database is represented as a n ×n matrix, where n is the number of pixels in an image, this matrix is then transformed to a vector of dimension nn ×1. Each image is transformed into these vector and a matrix M of these vectors of the images is formed where each column of M is a vector corresponding to an image in the