Medication Code Recognition using Convolutional Neural Network

Abstract

In this paper, a new automatic method for expiration code of medical products using convolutional neural network (CNN) is presented. The input image is enhanced using unsharp masking method. Then the image is binarized using local adaptive thresholding technique (LATT) and thinned using morphological operator. Also, characters of the image are extracted using bounding box technique. Finally, a set of characters (A-Z) and digits (0-9) is boiled. The dataset of characters feed an adopted architecture of CNN in order to recognize expiration code of the medication. The proposed approach is tested on large datasets of characters under various conditions of complexities. The experimental results demonstrate the robustness of our approach. The developed system achieves approximately 93% accuracy on character recognition.