Logo classification of amphetamines by surf and bag-of-features model

Abstract

An amphetamine is one of the drugs apprehended by the policemen in Thailand. At present, when an amphetamine drug seller is arrested, captured amphetamines are sent to the Scientific Crime Detection Center Region 1 to identify the amphetamines' source. Each drug is classified based on the printed character to relevant information from the source of drugs. For each case, a large volume of drugs is sent to be classified by only three staff members. It is a time-consuming task. In this work, we propose a framework for classifying the image of amphetamines based on their logo using the SURF and Bag-of-features model (BoF). In this work, we have dataset consists of three types: Apple logo for 192 images, R logo for 103 images, and WY logo for 360 images. We found that the unsmooth surface and low contrast are the main factors of low accuracy for this classification. Therefore, we propose a process to enhance the main feature and reduce noise on the surface using an adaptive filter, Contrast-Limited Adaptive Histogram Equalization (CLAHE), active contour, and image morphology. Our proposed algorithm shows that the clarity of the logo on amphetamines is improved. We also then apply SURF to extract features and classify using BoF. This experimental result shows for each step can improve the accuracy and the accuracy of our method is up to 97 percent. The accuracy of the WY logo classification by our proposed method is 94 percent.