Medicine Tablets Identification Using Feature Extraction Based on Neutral Network

Abstract

The purpose of this research is to develop medicine tablet identification method by using an image recognition system. The algorithm consists of three main stages: (i) preprocessing and segmentation (ii) feature extraction (iii) image classification. In the first stage, the images of the medicines are cropped. Then, we use Canny’s algorithm to detect edges from the images. After that in the second stage, 3 groups of features: shape features (perimeter, area, compactness, radius, standard derivation of radius), color features and internal tablet features are extracted from the medicine tablet images. In the last stage, we apply feed-forward backpropagation neural network to classify different groups of shape features. With these outputs, we compare and select the group of features that give the best classification. Finally, medicine tablet images are identified by using output from the neural network and observing color and internal tablet features.

            In the experiments we used 33 types of medicine, 20 tablets for each type which were digitized in 2 dimensions: front and back for each tablet. Each type of medicine was divided to 2 groups: training and testing groups equally. The proposed method yields maximum 99.39% total accuracy.

Keywords: Medicine Tablet Identification, Feature Extraction, Neural Network

Corresponding author: E-mail: jara838chan@yahoo.com, jaravee_ch@ccpe.or.th