Offline Handwritten Thai Character Recognition Using Single Tier Classifier and Local Features

Abstract

Handwritten character recognition is a conversion process of handwriting into machine-encoded text. Currently, several techniques and methods are proposed to enhance accuracy of handwritten character recognition for many languages spoken across the globe. In this project, a local feature-based approach is proposed to enhance the accuracy of handwritten offline character recognition for Thai alphabets. In the experiment, through MATLAB, 100 images for each class of Thai alphabets are collected and k-fold cross validation is applied to manage datasets to train and test. A gradient invariant feature set consisting of LBP and shape features is extracted. The classification is operated by using query matching based on Euclidean distance. The accuracy would be the percentage of correct classification for each class. For the result, the highest accuracy is 68.96% which has 144-bit shape features and uniform pattern LBP for the features.