Bayesian labelling of image corner features using a grey-level corner model with a bootstrapped modular neural network

In this paper we report the results of Bayesian labelling corner features in images with a modular neural network trained on data from a grey-level model of corners. Since adequate learning the whole mapping by a single network is problematic we present a bootstrapping procedure to partition data across modules. Results on real images are presented together with comparisons with the same labelling task performed by a monolithic network.