Recognition of Hand written Numerals on bank Cheques using Neural Networks

Abstract

People have often attempted to create computers that could perform human-like tasks. Man has been very effective in reducing the amount of physical labour required to perform many tasks by developing machines. With the invention of computers, it became possible for computers to minimize the number of workers required for a variety of jobs. However, there is one human skill that computers have yet to master. That is, the ability to recognize a person's handwriting. Of course, advances in machine handwriting recognition have been made, few of us imagine, moreover, that a computer would ever be able to read a human's handwriting or recognize a human's image profile Even so, approach that focuses that can imitate human recognition abilities, it is not a bad idea. After all, even imperfect handwriting leaves an impression. This paper investigates the problem of reading the numerical amount sector. In the case of cheques, segmenting unconstrained strings into individual digits is problematic due to joined and overlapping digits, disconnected digits, and digits that are physically attached to bits of digits from neighboring digits. There are five steps to the proposed architecture: block of courtesy number locator, segmentation of the string into individual digits, normalization, Gabor filter feature extraction, and neural network classifier recognition of each character.