Multi-Style Printed Sinhala Character Recognition and Digitalization Using Artificial Neural Network

Sinhala is the main language that is being used by over 75% of people as the first language and nearly 20% of people as a second language. Also, the Sinhala script is used to communicate and do official and non-official works all over Sri Lanka. Because of the technology developed in the world new trend was begin that digitalize printed documents because digitalizing documents are safe, easy to find, and can create editable files easily. Recognizing Sinhala characters are more complex mainly because of the shape and the character similarity. When digitalizing Sinhala printed documents using machine-trained models, it’s hard to detect similar characters with multi-style. Past researches are also done for printed Sinhala character recognition but for the training and testing, they used regular font styles only. The problem of those researches is whether the recognition was successfully done or not when testing special font styles like bold, italic, and underline characters. The objective of this research is to recognize multi-style printed Sinhala characters with their features using an Artificial Neural Network (ANN) based on Universe of Discourse and Self Organization Map methods. The developed model is tested with MATLAB 2015. Overall character recognition accuracy is over 85%. Normal Sinhala character recognition accuracy is the same as previous researches (over 90) but fonts with feature recognition accuracy levels have been increased over 80%.