International Conference on Advances in Pattern Recognition

ICAPR 2001: Advances in Pattern Recognition — ICAPR 2001 pp 83-92

Invariant Face Detection in Color Images Using Orthogonal Fourier-Mellin Moments and Support Vector Machines

  • Terrillon Jean-Christophe 
  • Mahdad N. Shirazi
  • Daniel McReynolds
  • Mohamed Sadek
  • Yunlong Sheng
  • Shigeru Akamatsu
  • Kazuhiko Yamamoto
Conference paper

DOI: 10.1007/3-540-44732-6_9

Volume 2013 of the book series Lecture Notes in Computer Science (LNCS)
Cite this paper as:
Terrillon JC. et al. (2001) Invariant Face Detection in Color Images Using Orthogonal Fourier-Mellin Moments and Support Vector Machines. In: Singh S., Murshed N., Kropatsch W. (eds) Advances in Pattern Recognition — ICAPR 2001. ICAPR 2001. Lecture Notes in Computer Science, vol 2013. Springer, Berlin, Heidelberg

Abstract

This paper proposes an automatic face detection system that combines two novel methods to achieve invariant face detection and a high discrimination between faces and distractors in static color images of complex scenes. The system applies Orthogonal Fourier-Mellin Moments (OFMMs), recently developed by one of the authors [1], to achieve fully translation-, scale- and in-plane rotation-invariant face detection. Support Vector Machines (SVMs), a binary classifier based on a novel statistical learning technique that has been developed in recent years by Vapnik [2], are applied for face/non-face classification. The face detection system first performs a skin color-based image segmentation by modeling the skin chrominance distribution for several different chrominance spaces. Feature extraction of each face candidate in the segmented images is then implemented by calculating a selected number of OFMMs. Finally, the OFMMs form the input vector to the SVMs. The comparative face detection performance of the SVMs and of a multilayer perceptron Neural Network (NN) is analyzed for a set of 100 test images. For all the chrominance spaces that are used, the application of SVMs to the OFMMs yields a higher detection performance than when applying the NN. Normalized chrominance spaces produce the best segmentation results, and subsequently the highest rate of detection of faces with a large variety of poses, of skin tones and against complex backgrounds. The combination of the OFMMs and of the SVMs, and of the skin color-based image segmentation using normalized chrominance spaces, constitutes a promising approach to achieve robustness in the task of face detection.

Keywords

Automatic face detection Skin color-based image segmentation Invariant moments Support vector machines Multilayer perceptron 

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

Copyright information

© Springer-Verlag Berlin Heidelberg 2001

Authors and Affiliations

  • Terrillon Jean-Christophe 
    • 1
  • Mahdad N. Shirazi
    • 2
  • Daniel McReynolds
    • 3
  • Mohamed Sadek
    • 4
  • Yunlong Sheng
    • 3
  • Shigeru Akamatsu
    • 4
  • Kazuhiko Yamamoto
    • 1
  1. 1.Office of Regional Intensive Research ProjectSoftopia Japan FoundationOgaki-City, GifuJapan
  2. 2.Communications Research LaboratoryKansai Advanced Research CenterKobeJapan
  3. 3.Centre d’Optique, Photonique et Laser, Département de PhysiqueUniversité Lavalste-FoyCanada
  4. 4.ATR Human Information Processing Research LaboratoriesKyotoJapan