Online character recognition

Abstract

Data entry using a pen forms a natural, convenient interface especially for handheld devices, which are very common now. The large number of writing styles and the variability between them makes the problem of writer-independent handwriting recognition a challenging pattern recognition problem. The structural approaches has long been dominating the online character recognition (OLCR) technology, in which the structure of input character is extracted and matched with the structure of models already stored in a model database to determine the class of input character. In this paper, we extracted the structure of a character as a sequence of primitives with their direction information as it is most widely used technique in OLCR research. In this approach, it is possible to have the same sequence of primitives for many different characters, especially for writer-independent environment introducing ambiguity in recognition of such characters. We have introduced a novel concept of 'relative connectivity' among the subsequent primitives to remove the ambiguity between different characters having the same sequence of primitives. Our observations show that almost all ambiguities in which different characters were having same sequence of primitives have been removed by using this novel 'relative connectivity' approach. As this technique succeed in removing ambiguities, very good recognition results, 98.3% for digits and 99.2% for uppercase letters, also been observed.