Implications of physical symmetries in adaptive image classifiers

It is demonstrated that rotational invariance and reflection symmetry of im age classifiers lead to a reduction in the number of free parameters in the classifier. When used in adaptive detectors, e.g. neural networks, this ma y be used to decrease the number of training samples necessary to learn a g iven classification task, or to improve generalization of the neural networ k. Notably, the symmetrization of the detector does not compromise the abil ity to distinguish objects that break the symmetry. (C) 2000 Elsevier Scien ce Ltd. All rights reserved.