A novel off-line character recognition: an MLP approach

Abstract

The purpose of this thesis work is to explore the possibility of efficient man-machine communication through printed documents. An attempt has been made to show the pattern recognition techniques i.e., KNN classifier helpful in recognition of machine printed characters and artificial neural networks may be used to represent and recognize printed English characters of any font and size. In our current work the machine printed document images are scanned by a front end video scanner and are applied to noise removal techniques using smoothing and sharpening filters. The noiseless images are digitized into a bi-level image using Ni-Black proposed binarization technique and proposed adaptive thresholding algorithm using Laplacian sign. Our work is split into three parts. The first part deals with segmentation and thinning. The output of this phase is thinned character image. The second part involves features are extracted from thinned image. The third part deals with KNN classifiers and training of the multilayer perceptron and recognizing characters after the system is trained. Automatic character recognition system promises to hold great future in Automatic office information processing system by integrating with multimedia, like Graphics, image and voice, into a single work station.