Segmentation of overlapping Sinhala handwritten characters

Abstract

Sinhala is the official and national language of Sri Lanka. Seventeen million people of Sri Lanka use Sinhala language to their day to day works. Most of the researches have been done to Sinhala printed character recognition with high accuracy. Nowadays, Sinhala handwritten character recognition is popular research in Sri Lanka. It is not like printed character segmentation; shape of the same type of handwritten character can be changed in different times. Therefore, characters will be overlapped or touched with each other. Handwritten character segmentation is more important to increase the accuracy of the character recognition. Currently there is lack of high accuracy finding to segment overlapping and touching Sinhala handwritten characters. The proposed methodology has six main sections. They are image acquisition, preprocessing, segmentation, classification, feature extraction and recognition. Collected image was loaded to the system and preprocessed it. Preprocess section is included noise removing, thresholding etc. After that, text line segmentation was done using horizontal projection profile. Mainly this research was introduced a connected pixel labeling method to segmentation of overlapping characters and peak and valley point identification method to segmentation of touching characters. According to tested result, connected pixel labelling method has 97% accuracy and peak and valley identification method has 72% accuracy.