A Back Propagation Based Face Recognition Model, using 2D Symmetrical Gabor Features

We present a system for recognizing human faces from a database consisting of multiple images per test subject, which spans the normal variations in a human face. The faces are represented based on a Gabor wavelet transform. The features are extracted as a vector of values using a carefully chosen symmetrical Gabor wavelet matrix. This feature extraction is biologically motivated and models systems based on human vision. The extracted features are fed into an artificial neural network, in dual phases. The training and testing phases of the neural network work on the features extracted by the same method. Excellent pattern-recognition-specific neural network like a multilayer perceptron with back propagation provides the necessary classification once the feature extraction is complete