Online Handwritten Character Recognition Using Wavelet-Based Features

Abstract

In today’s era, everyone is busy to the fullest and with increasing digitization everyone wants one’s work to be completed easily and as fast as possible. If it was possible to give input to a digital machine in our own handwriting and in our own native language, it would have been easier and probably less time consuming. Researchers are continuously working on handwriting recognizers. One of the basic steps in building a handwriting recognizer is recognition of handwritten characters. Handwritten Character Recognition can be done in two ways, namely, Offline and Online. This dissertation deals with the recognition of Online Handwritten Punjabi Characters. The steps involved in this process are capturing the inputs, preprocessing the data, extracting the features and finally the classification or recognition of the character captured. Feature extraction is a very important step in this process. A number of extraction methods have been proposed in literature for this process. This dissertation targets this field of research in the Punjabi language. Inputs were collected from various users in their handwritings using digital input. The Punjabi Character or Numeral taken as input was then converted to an image. This image was preprocessed to bring it to a standard size. Three different sizes have been tested in this work. The features have been extracted through wavelet decomposition of the preprocessed images. The features thus obtained are the wavelet coefficients. These coefficients have been used to train the SVM classifier to get the accuracy of the overall recognition system, i.e., percentage of correct classification or recognition of the character or numeral given as input. Complete methodology of the system and its overall performance and effectiveness has been demonstrated with the help of suitable examples.