A Decision Tree based Font Style/Size Independent Kannada Printed Character Recognition System

Segmentation is one of the most critical as well as important stage in optical character recognition system. Especially the segmentation of South Indian scripts has become one of the challenging aspects in order to provide a standard solution to South Indian OCRs. The segmentation of Kannada and Telugu scripts are considered to be still more serious researches due to the highest number of characters and increased variability, touching characters and overlapping characters in its native characters. This paper aims at providing an efficient touching line segmentation and classification algorithm in application with multiple projection profiles, bounding box analysis, Pearsons correlation features and decision tree classifier. The algorithm has provided improved accuracy in recognizing the complex or overlapping characters and proved to be efficient by obtaining around 97 % - 99 % of accuracy.