Development of Serial Number Extractor for Nigerian Currencies

Abstract

Abstract

The need to track currency movement and validate currencies in circulation are two cogent reasons for developing a real-time currency serial number extractor. Due to significant technical innovation over the past few decades, currency counterfeiting issues have gotten progressively worse all over the world and it is presently one of the main issues in Nigeria. Hence, financial institutions and the general public often desire to know the authenticity of cash. Also, Government and security agencies often desire to track cash for the purpose of apprehending notorious kidnappers after ransom collection. This calls for a fast and accurate serial number extraction from currencies. Automatic serial number extraction can be segmented into three (3) phases, which include currency classification, region of interest extraction, and character recognition. In this paper, an optimized object identification model was proposed for currency classification. A pre-trained denomination-based region of interest algorithm was then applied for the extraction of the serial number region. We further utilized three (3) existing optical character recognition models: Pytesseract, Easy OCR, and Keras OCR for the character recognition. Accuracy, Levenshtein Distance, Jaccard Similarity, Character Error Rate, and Damerau Distance metrics were employed in this study for recognition performance measure. The currency bill classification yielded a 100% accuracy while the best accuracy of 81% was obtained with the EasyOCR framework at the character recognition phase. The recognition model performance can be considered poor for the targeted application. Hence, the need for customization of the general character recognition architecture for Nigerian currency bill serial number recognition.