A Comparative Study of Face Representations in the Frequency Domain

  • Eduardo Garea Llano
  • Josef Kittler
  • Kieron Messer
  • Heydi Mendez Vazquez
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 4225)


The variation of illumination of an object can produce large changes in the image plane, significantly impairing the performance of face verification algorithms. In this paper we present a comparison of several face representation methods from the point of view of their sensibility to illumination changes. The sensibility is measured in term of the overlap of distribution of normalized correlations for inter class and intra class image comparison. We compared a combination of differentiated image in the frequency domain and the performance of Fourier parameters to obtain an illumination insensitive representation. The result suggests, that better illumination invariance could be achieve in feature spaces developed for a differentiated image rather than using the original input image.


  1. 1.
    Acton, S.: Multigrid Anisotropic Diffusion. IEEE Trans. Image Processing 7 (1998)Google Scholar
  2. 2.
    Adini, Y., Moses, Y., Ullman, S.: Face recognition: the problem of compensating for illumination changes. IEEE Trans. Pattern Anal. Mach. Intelligence, 721–732 (1997)Google Scholar
  3. 3.
    Bailly-Bailliere, E.: The BANCA Database and Evaluation Protocol AVBPA (2003)Google Scholar
  4. 4.
    Belhumeur, P., Hespanha, J., Kriegman, D.: Eigenfaces vs. Fisherfaces: Recognition using Class Speci.c Linear Projection. IEEE Trans. Pattern Anal. Mach. Intelligence 19, 711–720 (1997)CrossRefGoogle Scholar
  5. 5.
    Turk, M., Pentland, A.: Eigenfaces for Recognition. J. Congitive Neuroscience 3, 71–86 (1991)CrossRefGoogle Scholar
  6. 6.
    Belhumeur, P., Hespanha, J.P.: Eigenfaces vs. Fisherfaces: Recognition using class speci_c linear projection. IEEE PAMI 19, 711–720 (1997)Google Scholar
  7. 7.
    Belhumeur, P., Kriegman, D.: What is the set of images of an object under all possible lighting conditions. Int. J. of Computer Vision 28, 245–260 (1998)CrossRefGoogle Scholar
  8. 8.
    Georghiades, A., Kriegman, D., Belhumeur, P.: From few to many: Generative models for recognition under variable pose and illumination. IEEE PAMI (2001)Google Scholar
  9. 9.
    Riklin-Raviv, T., Shashua, A.: The Quotient image: class-based re-rendering and recognition with varying illumination conditions. In: IEEE PAMI (2001)Google Scholar
  10. 10.
    Lee, K., Ho, J., Kriegman, D.: 9 Points of Light: Aquiring Subspaces for Face Recognition Under Variable Lighting. In: IEEE Proc. Conf. Computer Vision and Pattern Recognition (2001)Google Scholar
  11. 11.
    Der Wei, S., Lai, S.-H.: Robust Face Recognition under lightinng Variations. In: Proc. Conf. on Pattern Recognition IEE (2004)Google Scholar
  12. 12.
    Barnamas, T.: Comparing face images using the modified Hausdorff distance. Pattern Recognition 31(12), 1873–1881 (1991)Google Scholar
  13. 13.
    Gao, Y., Leung, K.: Face recognition using line edge map. IEEE Trans. Pattern Analysis Mach. Intel. 24(6), 764–779 (2002)CrossRefGoogle Scholar
  14. 14.
    Savvides, M., Vijaya Kumar, B.V.K., Khosla, P.K.: Eigenphases vs. Eigenfaces. In: Proc. Conf on Pattern Recognition, IEEE, Los Alamitos (2004)Google Scholar
  15. 15.
    Luettin, J., Maitre, G.: Evaluation Protocol for the XM2FDB Database In: IDIAP 1998 (1998)Google Scholar
  16. 16.
  17. 17.
    Short, J., Kittler, J., Messer, K.: A comparison of photometric normalization algorithms for face verification. In: Kanade, T., Jain, A., Ratha, N.K. (eds.) AVBPA 2005. LNCS, vol. 3546, pp. 617–626. Springer, Heidelberg (2005)CrossRefGoogle Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2006

Authors and Affiliations

  • Eduardo Garea Llano
    • 1
  • Josef Kittler
    • 2
  • Kieron Messer
    • 2
  • Heydi Mendez Vazquez
    • 1
  1. 1.Advanced Technology Application CenterSiboney PlayaCuba
  2. 2.Centre for Vision Speech and Signal ProcessingUniversity of SurreyUK

Personalised recommendations