Bio-Inspired Adaboost Method for Efficient Face Recognition

We present the design of face recognition system based on the Adaboost algorithm and bio- inspired evolutionary search. We start by extracting the feature vector of the face image based on fixed fiducial points. Then we decompose the strong feature into several feature subsets using GA and classification models of each feature subsets are combined using the Adaboost algorithm. GA searches the best feature combination that gives minimum training error. We use the fixed feature decomposition method, where the length of the feature subset is constant. We use Gabor filter of 8 orientations and 8 frequencies to extract the feature of the face. Experiments are conducted on FERET database which contains 2418 images of 1209 subjects taking 2 images per subject. The outcome of these experiments suggests that the classification model using aggregation of feature combinations by means of Adaboost and GA gives better result than classification model that uses the entire feature vector.