Genetically evolved transformations for rescaling online handwritten characters

In this paper, we describe using genetically evolved rescaling transformations for scale normalization of online handwritten characters. GP is used for evolving the rescaling functions such that the transformed characters make less recognition errors. The procedure for evolving these transformations requires only labeled characters for training and testing the classifier, but no knowledge of the internals of the classifier. It therefore holds promise as a general mechanism for improving the recognition performance of any online HWR algorithm.