Multiresolution locally expanded HONN for handwritten numeral recognition

Abstract

In this paper, we propose a neural network architecture, multiresolution locally expanded high order neural network (MRLHONN) to solve the problem of handwritten numeral recognition. In this recognition scheme, the multiresolution representation of character image is input into a high order neural network (HONN), while in each resolution, only neighboring pixels are expanded to produce high order input. The property of this architecture is that, the local expansion alleviate the problem of large connecting weight set, and the multiresolution representation remedy the inadequacy of local expansion. Two forms of multiresolution representations, quadtree representation and Gaussian pyramid, were used in experiments. The recognition results demonstrate the efficiency of the proposed architecture.