Learning compatibility functions for feature binding and perceptual grouping

Abstract

We present and compare data driven learning methods to generate compatibility functions for feature binding and perceptual grouping. As dynamic binding mechanism we use the competitive layer model (CLM), a recurrent neural network with linear threshold neurons. We introduce two new and efficient learning schemes and also show how more traditional standard approaches as MLP or SVM can be employed as well. To compare their performance, we define a measure of grouping quality with respect to the available training data and apply all methods to a set of real world fluorescence cell images. © Springer-Verlag Berlin Heidelberg 2003.