A writer adaptation method for isolated handwritten digit recognition based on Ensemble Projection of features

Learning handwriting categories fail to perform well when trained and tested on data from different databases. In this paper, we propose a novel framework of Ensemble Projection (EP) for writer adaptation. We employed EP as a feature transformation method which can be combined with different types of classifiers for unsupervised and semi-supervised adaptation. Experiments on a handwritten digit dataset demonstrate that EP learning can increase recognition rates significantly, both in the unsupervised and semi-supervised cases.