On Segmentation of Words from Online Handwritten Gurmukhi Sentences

Abstract

Keyboard based devices face many problems relating to hardware failure or damages occurring due to multiple users handling a device and decaying of the device as hardware gets old. So there is a need to provide other ways of communication between machine and human beings, it is done through speech input and handwriting input. Handwriting is a natural way of communication between machine and human with pen-based technology emerging rapidly. This thesis deals with word segmentation from online handwritten Gurmukhi sentences. For recognizing sentences, proper segmentation of sentences into words is very important. The method proposed in this thesis for word segmentation considers online data at stroke level in which white spaces between the words are not explicitly known. This work, as such, focuses on the recognition of online Gurmukhi sentences. Thresholding approach is used for segmenting words from a sentence. The basic handwriting unit of a sentence is a stroke. One or more strokes form the words and creation of words is nothing but a sentence. In the proposed approach, vertical gap between the strokes are first located and then based on the maximum threshold value the word is extracted from the sentence. Testing of proposed approach has been performed on 200 sentences. A segmentation accuracy of 91.0% has been achieved with the use of the proposed algorithm for segmenting sentence into words for the online Gurmukhi script.