Pill Identification with Imprints Using a Neural Network

doi: 10.14456/mijet.2015.7

Abstract

Since there are more than ten thousand different types of pills commonly used in hospitals in Thailand, they are not easily recognized by an inexperienced pharmacist or even an experienced pharmacist. Thus, our long term goals are to develop a commonly-used pill database in Thailand and to build a system to assist pharmacists to identify unknown pills in real-time. In this paper, we focused only on using imprints to identify pills that are almost identical in color and shape. We proposed a new algorithm to extract the feature vector from the imprints. The extracted feature vector was invariance to image rotation and was fed into the neural network. The neural network was used to identify the type of an unknown pill. Although the number of pill types was limited to six in this work, the results were promising. The percent accuracy was about 94.4%.