Pill image binarization for detecting text imprints

Abstract

Identifying pill is a major concern for both patients and pharmacists nowadays. Being able to extract pill information automatically will benefit both pill information retrieval and indexing processes. We aim to extract texts from pill imprints so that they can be used to search existing pill databases. This paper presents approaches for extracting texts from pill images based on i) a technique presented by Kasar[1] and ii) processing edge masks of imprints. We also compared different thresholds for binarizing extracted text area so that it can be used with optical character recognition (OCR). The result showed that the method based on using edge mask performed better, and the Otsu threshold gave the best results for binarization of the imprint area.