Online Sinhala Handwritten Character Recognition on Handheld Devices

Abstract

The online handwritten character recognition on handheld devices presents a challenging assignment due to the limited memory, processing and storage constraints. This paper presents an online handwritten character recognition process based on Non Deterministic Finite Automaton (NFA) and XML character pattern profiles for Sinhala alphabetic characters. The process models a character as a sequence of directional states and accumulates under a hierarchical XML tag structure. The NFA is used to construct valid Sinhala character combinations and to determine pattern search alternatives. Accuracy improvement and failure analysis models are used to achieve a recognition rate of 90% on run on discretely written Sinhala characters after adopting user specific customizations on character pattern profiles.