Phase-eigen subspace based illumination invariant face recognition using associative memory

Phase eigen subspace based face recognition under varying lighting conditions is proposed. Universal subspace analysis is exploited in frequency domain and phase spectrum is extracted instead of using raw spatial data of face images. Improved results are obtained when simplified bi-directional associative memory neural network is used as classifier. The proposed scheme is experimented over two standard databases like YaleB and PIE and the promising recognition accuracy is achieved while comparing to standard subspace methods.