Development of a learning algorithm for facial recognition under varying illumination

Face recognition under varying illumination and dimensionality reduction has been a key problem in the field of Computer Vision. An extension of Principal Component Analysis (PCA) called Independent Component Analysis (ICA) has been utilised in this paper as a feature extraction technique. In the proposed approach three feature selection techniques have been investigated namely Adaboost, Gentle-Adaboost and Sequential Forward Floating Selection (SFFS). The classifier used in this research is the supervised learning algorithm, Support Vector Machines (SVM). The algorithm was tested on the Extended Yale B face database and a five person database created for the purpose of this research. Face recognition rate was highest when all features extracted were used for recognition and when feature selection techniques Adaboost and Gentle Adaboost were used. Therefore, we conclude that by using these two feature selection techniques only 23% of the total features are required giving the same accuracy as when using all features. This increases dimensionality reduction and performance of the algorithm enabling recognition of frontal face images under varying illumination.