LCN - Laboratory of Computational Neursocience

Support Vector Machines and Saccadic Search

The Gabor Decomposition is a classical tool in Computer Vision. However, Gabor features are computationally expensive and are therefore often considered ill-suited for active-vision applications. The Saccadic Search algorithm we propose is a bio-inspired approach to fast Gabor Decomposition based pattern recognition.

The limitations of a common, plain Gabor algorithm can be overcome by integrating two main inputs from biology, namely

The human saccadic system
The arrangement of photoreceptors in the retina

The integration of these two concepts into the common Gabor algorithm constitutes the Saccadic Search strategy.

A few applications of Saccadic Search are presented here (for more details, please have a look at our publications online):

Support Vector Machines (SVM) are powerful classifiers with good generalization properties. In contrast to many ad-hoc methods, SVMs have a solid theoretical foundation in machine learning and statistical pattern recognition. We use SVM's in the classification steps of saccadic search, recognition, and authentication.

Facial Feature detection (this also employs SVMs)
Face Authentication (SVMs also used here)
Real-Time Head Detection and Tracking (Features a novel eyeball-mounted camera developed by the Biovision and Robotics Group)

The Facial Features detection and Face Authentication algorithms have also been implemented in the VIRSBS demonstrator.

Collaborators:

Some recent publications:

Face Authentication by retinotopic sampling of the Gabor decomposition and Support Vector Machines
F. Smeraldi, N. Capdevielle, and J. Bigun,
accepted for publication in the Proceedings of the 2nd International Conference on Audio and Video Based Biometric Person Authentication (AVBPA'99), Washington DC (USA), March 22nd-24th, 1999
View Abstract - Download PostScript

Saccadic Search with Gabor features applied to Eye Detection and Real-Time Head Tracking,
F. Smeraldi, O. Carmona and J. Bigün. accepted by Image and Vision Computing, 1998
View Abstract - Download PostScript

Facial feature detection by Saccadic Exploration of the Gabor Decomposition
F. Smeraldi and J. Bigun,
Proceedings of the 1998 International Conference on Image Processing, Chicago (USA), vol. 3, pages 163 - 167, October 4 - 7, 1998
View Abstract - Download PostScript

Real-time Head Tracking by Saccadic Exploration and Gabor Decomposition
F. Smeraldi, O. Carmona and J. Bigun,
Proceedings of the 5th International Workshop on Advanced Motion Control (AMC '98), Coimbra (Portugal), June 29th - July 1st, pages 684 - 687, IEEE Cat. Num. 98TH8354, 1998
View Abstract - Download PostScript

Please send comments on this page to: [email protected]

Go to LCN - DI - EPFL

LCN - Laboratory of Computational Neuroscience