Learning banknote fitness for sorting

In this work, a machine learning method is proposed for banknote soiling determination. We apply proven techniques from computer vision to come up with a robust and effective method for automatic sorting of banknotes. The proposed method is evaluated with respect to various invariance classes. The method shows excellent performance on a large validation set of over 8,000 banknotes from the Eurosystem, while being learned on only 300 banknotes per denomination.