On the recognition of offline handwritten word using holistic approach and AdaBoost methodology

Offline handwritten word recognition assumes an imperative part in the domain of document analysis and recognition. This article describes a technique for the recognition of offline handwritten Gurumukhi words. The proposed system uses a holistic approach to recognize a word, where a word itself is considered as an individual item. Thus, the word is recognized without considering any explicit segmentation. A set of features, i.e. zoning features, diagonal features, intersection & open-end point features is considered to extract the desirable characteristics from the word images. The classification techniques like k-Nearest Neighbor (k-NN), Support Vector Machine (SVM) and Random forest classifiers are employed for the recognition purpose. To boost the system performance, majority voting scheme of all the considered classifiers and an ensemble algorithm i.e. AdaBoost (Adaptive Boosting) algorithm are used. This system is evaluated on the database comprising 1,00,000 samples of 100 different city names handwritten in Gurumukhi script. Maximum recognition accuracy of 88.78% has been achieved using AdaBoost methodology and the attained results are comparable with state-of-the-art results.