Banknote Issuing Country Identification Using Image Processing and Neural Networks

摘要

The work in this paper investigates developing an identification system for 21 countries using images of their banknotes and neural network classifiers. We consider the banknotes of 19 Asian countries, the European Union (EU), and the USA. Our motivation to investigate the Asian currencies is the increased global interaction in tourism and international trading with these countries where they have diverse and impressive banknote designs; thus making it difficult to identify by foreign visitors or traders. Our database comprises 504 original and pre-processed images of 6 banknotes of each of the 21 currencies. The investigated 19 Asian countries in this work are Afghanistan, Armenia, Azerbaijan, Bangladesh, Bhutan, Brunei, Burma, Cambodia, China, India, Kuwait, Maldives, Pakistan, Saudi Arabia, Sri Lanka, Syria, Tajikistan, Turkey, and United Arab Emirates. Most existing banknote identification systems aim to identify the currency value or decide whether a banknote is counterfeit. Our presented work is novel as it focuses on identifying the issuing country. Furthermore, we apply two pattern-averaging methods using (5×5) and (10×10) kernels, and follow two learning schemes to train and test the proposed neural identification models by using (50:50) and (75:25) training-to-validation data ratios. The obtained experimental results are considered as successful.