Gabor wavelets and General Discriminant Analysis for face identification and verification

A novel and uniform framework for both face identification and verification is presented in this paper. The framework is based on a combination of Gabor wavelets and General Discriminant Analysis, and can be considered appearance based in that features are extracted from the whole face image. The feature vectors are then subjected to subspace projection. The design of Gabor filters for facial feature extraction is also discussed, which is seldom reported in the literature. The method has been tested extensively for both identification and verification applications. The FERET and BANCA face databases were used to generate the results. Experiments show that Gabor wavelets can significantly improve system performance whilst General Discriminant Analysis outperforms other subspace projection methods such as Principal Component Analysis, Linear Discriminant Analysis, and Kernel Principal Component Analysis. Our method has achieved 97.5% recognition rate on the FERET database, and 5.96% verification error rate on the BANCA database. This is a significantly better performance than that attainable with other popular approaches reported in the literature. In particular, our verification system performed better than most of the systems in the 2004 International Face Verification Competition, using the BANCA face database and specially designed test protocols.