MODIFIED DIRECTION FEATURE AND NEURAL NETWORK BASED TECHNIQUE FOR HANDWRITING CHARACTER RECOGNITION

Abstract

Handwriting character recognition is an important research topic which has various applications in surveillance, radar, robot technology... In this paper, we propose the implementation of the handwriting character recognition using off-line handwriting recognition. The approach consists of two steps: to make thin handwriting by keeping the skeleton of character and reject redundant points caused by humam’s stroke width and to modify direction method which provide high accuracy and simply structure analysis method to extract character’s features from its skeleton. In addition, we build neural network in order to help machine learn character specific features and create knowledge databases to help them have ability to classify character with other characters. The recognition accuracy of above 84% is reported on characters from real samples. Using this off-line system and other parts in handwriting text recognition, we can replace or cooperate with online recognition techniques which are ususally applied on mobile devices and extend our handwriting recognition technique on any surfaces such as papers, boards, and vehicle lisences as well as provide the reading ability for humanoid robot.