Offline Handwritten Punjabi Character Recognition using Artificial Neural Networks

Abstract

Handwritten Character recognition has been one of the challenging computational processes with conventional digital computers. New computational fields like expert systems and artificial intelligence have provided some level of recognition of handwriting. Even then, there exists competition between accuracy and speed. Yet, the human mind can decipher handwritten code accurately and quickly. This is possible only because of the involvement of a densely connected neural network. The idea that a non-biological neural network can be utilized to solve human processes has been proposed in 1948 by Wiener. However, till recently the computational capability of digital computers did not allow even time scaled simulation of neural systems. The recent developments in VLSI technology (in terms of microprocessors and memory) have permitted simulation and possible real-time model execution of neural networks for engineering applications (like filtering, control, object recognition, etc). We extend on this thought to create an artificial neural network based software system for offline handwritten Gurmukhi character recognition. The system utilizes a back propagation algorithm for training of Network. The character is written offline in MS Paint and stored as 24-bit bmp format of size 32x 32 matrix size. Any end user can test additional characters, as per requirement, on the system.