A facial recognition technique employing Frequency Domain Thresholding and Quantization (FD-TQ)

A facial recognition technique employing Frequency Domain Thresholding and Quantization (FD-TQ) is presented in this paper. The algorithm has attractive properties with respect to storage requirements and computational complexity in both the training and testing modes, which make the technique particularly suitable for large data bases. The new algorithm is applied to ORL, Yale and FERET data bases. The experimental results confirm the significant reduction in the storage and computational requirements compared with recently reported techniques, without sacrificing the recognition accuracy.