Gaussian mixture models based on the frequency spectra for human identification and illumination classification

The importance of Fourier domain phase in human face identification is well-established Hayes et al., (1982). It therefore seems natural that identification tools based on phase features should be very efficient. In this paper we introduce a model-based approach using Gaussian mixture models (GMM) based on phase for performing human identification. Identification is performed using a MAP estimate and we show that we are able to achieve misclassification error rates as low as 2% on a database with 65 individuals with extreme illumination variations. The proposed method is easily adaptable to deal with other distortions such as expressions and poses, and hence this establishes its robustness to intra-personal variations. Finally, we demonstrate that GMM based on the Fourier domain magnitude is effective for illumination normalization, so that near perfect identification is obtained using the reconstructed illumination-free images.