Handwritten digits recognition using Hough transform and neural networks

Abstract

A system for handwritten digits recognition using the Hough transform and a neural network has been developed. The input to the system consists on a 128 × 128 image containing one handwritten digit. This input is processed using the Hough transform and fed into the neural network, which in turn performs the recognition task. In order to decrease the size of the input vector to the neural net and still preserve the most of the information contained in the Hough space, this is sampled in a non-uniform way. It is also made translation and scale independent. An 80% mean recognition rate was obtained using a Kohonen's self organized feature map testing 720 samples of digits written by 18 different persons.