Tablet identification using support vector machine based text recognition and error correction by enhanced n-grams algorithm

Abstract

Unidentified and/or misidentified tablets always present challenges to both patients and health care professionals alike. Consumption of these misidentified tablets often results in adverse drug reaction and sometimes may even cause ill health leading to death. Thus, identification of unknown tablets is an important task in the medical industry. This study proposes an algorithm that uses the text imprinted on the tablet images to identify unknown images. Text imprinted on pills often contains important information that can be used to identify tablets. In this study, multiple feature sets were extracted from tablet images. The proposed work first identifies the text region, from which the text is recognised using support vector machine classifier. A post processing algorithm based on the confusion model combined with n-grams is used to correct the substitution, insertion and deletion errors in the text recognised. Finally, the corrected text is matched with a template tablet database to identify similar tablets. Experimental results showed that the proposed system is efficient with respect to accuracy and can be safely used by both common public and health care professionals to identify tablets.