Study and Development of Handwritten Numeral Character Recognition

Abstract

Image processing is basically used to extract useful information from any input image. Recognition has a very important role in image processing. In this exploration work, we have concentrated on the recognition of the handwritten numeral characters. Neural Network is used for recognizing the different handwritten numerals. Our method comprises of three stages and they are pre-processing, training and recognition. Pre-processing stages includes removal of noise, binarization, re-scaling and finding the skeleton of an image. Skew correction is also used for segmenting the different characters in an image. In training stage we have used back propagation technique for recognizing different numeral characters. Different hidden layers are used while training to have better accuracy. Recognition stage recognizes the different characters in an image from the trained neural network. The above proposed system has been performed in Matlab. The system detects the numerals with an exactness in around 90-95%.It works well and has the similar accuracy in even twisted pictures or pictures having different size.