Data driven learning for feature binding and perceptual grouping with the Competitive Layer Model

The present work describes the treatment of several grouping or segmentation problems from the field of automatic image processing. It is motivated by natural grouping principles which can be observed in human visual perception, like the Gestalt laws of proximity, similarity and closture. These principles are implemented by pairwise interactions between elementary data structures in a recurrent neural network architecture, the so called Competitive Layer Model (CLM). The main result of the work is the development of an automatic learning method which extracts suitable interaction patterns from exemplary target groupings.

Level thesis:degree_level

thesis.doctoral

Grantor dc:publisher

Universität Bielefeld

Year

2006

Author dc:creator

• Weng, Sebastian

Repository record source_url

https://pub.uni-bielefeld.de/record/2305763

OAI identifier oai:identifier

oai:pub.uni-bielefeld.de:2305763