A spectral domain feature extraction algorithm for face recognition

In this paper, a frequency domain face recognition algorithm is proposed, which exploits the variation in local spectral features. Instead of performing the face recognition task by extracting features from the entire face image, an entropy-based band selection criterion is developed, which selects high-informative horizontal bands. Moreover, a local feature selection algorithm is introduced to capture the variation of the spectral features within these high-informative horizontal bands in detail. Magnitudes and frequencies corresponding to the dominant two-dimensional Fourier transform coefficients are proposed to be selected as features and shown to provide high within-class compactness and high between-class separability. Extensive experimentations have been carried out upon two standard image databases and the recognition performance is compared with some of the existing face recognition methods. It is found that the proposed method offers not only computational savings but also a very high degree of recognition accuracy.