An algorithm development for handwritten character recognition by personal handwriting identity analysis [PHIA]

Abstract

The algorithm for online handwritten character recognition, PHIA algorithm, is introduced. The algorithm uses a likelihood score computed by a small neural network from every symbol pair for various decisions. Scores are used to generate a relationship map (Rivals/Non-rivals) between each symbol pairs. The training data is added to the database if and only if the relationship with the training data is `rival' for all existing database samples that identifies the same symbol. In the recognition phase, a nearest neighbor search is applied. During the search, if we traverse to a node whose relationship to the input is `non-rival', we later skip all processes that would operate on that node's rivals. This optimizes the decision path for each of the individual and enhances the ability to learn new symbols effectively.