Handwriting recognition

Abstract
----------
Handwriting recognition is the problem of transforming a word from the iconic form of writing to its symbolic form. The problem of handwriting script recognition has received attention from a number of researchers in the past. In this work, one application of Neural Network called Neocognitron is used. Fukushima’s Neocognitron [1-5] has received attention over the past two decades as a partially shift invariant, distortion tolerant classifier [3]. This is one of the most complex artificial neural network structures to simulate.
We have showed that using Neocognitron it is possible to make the machine to learn the general shape of a pattern and then use that to recognize the same pattern when it sees it again. In this paper, we have reviewed Fukushima’s work and the Neocognitron’s ability to recognize handwritten Arabic numerals, even with considerable deformations in shape, is demonstrated using a Windows platform.
We implemented the supervised version of the network following several of the papers written about the Neocognitron during the 1980s.The purpose of our implementation was mainly to understand the functionality of the Neocognitron, not to have a very efficient ‘work horse’. After reading the papers listed in the ‘References’ and ‘Bibliography’, we decided to implement a network similar in structure to the network mentioned in [3]. In that paper Fukushima supervises the network to recognize numerals. The reason to use similar topology as the one described in [3] was that the dimensions were explicitly stated. So we stayed with the original 4 stages (i.e. modules, S-C layer combinations). [4] Specifies most of the equations used for our network. This paper does not specify any dimensions for the network, but it has a precise mathematical appendix.
In the present paper, Neocognitron has been implemented in Windows platform and is shown to have a great capability for visual pattern recognition.

Handwriting recognition is the problem of transforming a word from the iconic form of writing to its symbolic form. The problem of handwriting script recognition has received attention from a number of researches in the past. In this work one application of Neural Network is used. Using Neocognitron it is possible to make the machine to learn the general shape of a pattern and then use that to recognize the same pattern when it sees it again.