Automated detection and classification of counterfeit banknotes using quantitative features captured by spectral-domain optical coherence tomography

Counterfeiting of banknotes is still a severe crime problem in many countries. One of the most significant issue for solving the crime is to classify the counterfeit types and identify the sources. Most of the current methods to classify counterfeit banknotes rely on manual examination that is time-consuming and labor-intensive. Moreover, these methods only detect surface features which can be easily imitated through advanced printing technology. In this study, an automated method based on optical coherence tomography (OCT) and machine-learning algorithms was proposed to classify different types of banknotes based on the internal features. A spectral-domain OCT (SD-OCT) system was employed for sub-surface imaging and quantitative assessment of banknotes. A total of 29 Chinese 100-Yuan banknotes were collected, in which 4 of them were real and 25 of them were counterfeiting by three different printing processes. Each banknote was imaged 10 times in 3 distinct regions, which resulted in a dataset of 290 samples. Each sample was characterized by extracting 2 A-scan (OCT signal intensity along depth) based features and 14B-scan (cross-sectional OCT images) based features. Several machine-learning models, including logistic regression (LR), support vector machines (SVM), K-nearest neighbor (KNN) and random forest (RF), were built and optimized as the classifiers that were trained using 203 samples and applied to predict 87 testing samples. The best performance was achieved by SVM classifier in which the sensitivity of 96.55% and specificity of 98.85% were obtained in discriminating between authentic and counterfeit banknotes, and the sensitivity of 94.67% and specificity of 98.22% were obtained in predicting the types of counterfeit banknotes. These classifiers were also evaluated using the receiver operating characteristic (ROC) curves. To the best of our knowledge, this is the first study where A-scan and B-scan derived features from OCT images have been used for the detection and classification of different types of counterfeit banknotes.