Pharmaceutical Blister Pack Recognition using Local Features

In this paper, we propose a high-accuracy, highspeed method for recognizing pharmaceutical blister packs. In the proposed method, we first rank reference images of blister packs stored in a database for a given query image by using a simple voting process based on a nearest-neighbor search for local features. Next, we evaluate the results in ranking order by similarity of shape and color with reference to a template image for each reference image. In the evaluation process, we utilize an arrangement of local features to specify the area in the query image that is similar to the template image. By doing so, we reduce the computational time compared to simple template matching methods. To test our proposed method, we constructed a prototype and tested it at a pharmacy. As a result of this experiment, we demonstrate that the proposed method offers a practical method of recognizing blister packs accurately and quickly.