Session:

Image Processing and Computer Vision
Tuesday, March 02, 2004, 10.50 – 11.10

Session Chair:

D. Linkens

Paper Title:

Pulsed Neural Networks for Feature Detection using Dynamic Synapses

Author(s):

Dr A.Heittmann, Infineon Technologies AG, Munich
U. Ramacher, Infineon Technologies AG, Munich

Abstract:

Technical applications where scene analysis is involved rely on a robust feature extraction. Feature extraction based on neural principles traditionally emphasize the analogue representation of information as well as analogue information processing at least to the level where simple cells are modeled. In contrast to this we will show in this paper how principles of correlation and detection of correlation can be used to extract features up to a level of complexity of at least that of simple cells. Using only two data-driven adaption rules for connection weights dependent on membrane potentials as well as on dendritic current simple networks can be implemented to detect gradients of intensity. By replicating of those detectors and defining densitiy of connectivity to the receptors of the retina complex detector-profiles can be implemented.