Character recognition using neural networks

Abstract

It has been 50 years since the idea popped up that calculating systems can be made on the replica of the biological neural networks. Still, the development of this science branch made the improvement of these systems possible only in the last 25-30 years. Nowadays, neural computing is a very extensive, separate science. Its solid theory basis made it possible to use them to solve many kind of problems in artificial computing, thus improving the experience of the science. Neural networks are commonly used to solve sample-recognition problems. One of these is character recognition. The solution of this problem is one of the easier implementations of neural networks. With the help of Matlab's Neural Network Toolbox, we tried to recognize printed and handwritten characters by projecting them on different sized grids (5×7, 7×11, 9×13). The results showed that the precision of the character recognition depends on the resolution of the character projection. Also, we realized, that not every writing style can be recognized using the same network with the same precision. This shows that the variety of human handwriting habits can't be fully covered with one neural network.