Face Authentication by retinotopic sampling of the Gabor decomposition and Support Vector Machines

Abstract:

We describe a face authentication algorithm based on retinotopic sampling of the Gabor decomposition. A log-polar sampling grid is used to encode information from relevant regions of the face. Decision is based on a multi expert scheme, in which each expert independently authenticates the client based on a different facial feature. Implementation of the expert as Support Vector Machines is discussed. An alternative implementation using K-Nearest-Neighbours classifiers is presented for comparison.