LCE Kotisivu

S-114.V Correspondence-based face and object recognition


S-114.V Correspondence-based face and object recognition (1 cr)

Lecturer: Ph.D. Rolf P. Würtz
Ruhr-Universität Bochum, Germany

Examination: Lectures and computer exercise

Lectures: 26.5.,27.5.,28.5. 13-15 B317

In this lecture I start by presenting wavelet analysis from theoretical and practical aspects and showing the connection to modeling of the visual system. The Gabor wavelet transform is introduced as a model of simple and complex cells in the primary visual cortex. Based on the so defined visual features, the visual correspondence problem can be treated, which is central to many computer vision problems. Techniques are elastic graph matching, pyramid matching, and a dynamic neural network solution. These techniques can be successfully applied to face and object recognition. They must be accompanied by appropriately organized object (face) memory and prefiltering of the video stream. This leads to bunch graph matching, one of the most powerful methods for face recognition in the world today. Finally, I describe various further applications like the recognition of hand and facial gestures, analysis of medical images, and an integrated user-guided pick-and-place behavior of a humanoid robot. This robot also features tactile sensors, which shall be used to learn stable grips for objects. An important boundary condition for all applications is the learnability of the required representations.

Lecture slides:


Further information: Aki Vehtari

Tästä sivusta vastaa
Sivua on viimeksi päivitetty 18.06.2003