FACE RECOGNITION - A CONVOLUTIONAL NEURAL-NETWORK APPROACH

Faces represent complex multidimensional meaningful visual stimuli and developing a computational model for face recognition is difficult. W e present a hybrid neural-network solution which compares favorably wi th other methods. The system combines local image sampling, a self-org anizing map (SOM) neural network, and a convolutional neural network. The SOM provides a quantization of the image samples into a topologica l space where inputs that are nearby in the original space are also ne arby in the output space, thereby providing dimensionality reduction a nd invariance to minor changes in the image sample, and the convolutio nal neural network provides for partial invariance to translation, rot ation, scale, and deformation. The convolutional network extracts succ essively larger features in a hierarchical set of layers. We present r esults using the Karhunen-Loeve (KL) transform in place of the SOM, an d a multilayer perceptron (MLP) in place of the convolutional network. The KL transform performs almost as well (5.3% error versus 3.8%). Th e MLP performs very poorly (40% error versus 3.8%). The method is capa ble of rapid classification, requires only fast approximate normalizat ion and preprocessing, and consistently exhibits better classification performance than the eigenfaces approach on the database considered a s the number of images per person in the training database is varied f rom one to five. With five images per person the proposed method and e igenfaces result in 3.8% and 10.5% error, respectively. The recognizer provides a measure of confidence in its output and classification err or approaches zero when rejecting as few as 10% of the examples, the u se a database of 400 images of 40 individuals which contains quite a h igh degree of variability in expression, pose, and facial details. We analyze computational complexity and discuss how new classes could be added to the trained recognizer.