Convolutional Neural Network Based Serial Number Recognition Method for Indian Rupee Banknotes

摘要

The recognition of the banknote serial number, which constitutes important data used for various purposes is one of the important functions of banknote counters. However, traditional character recognition methods are limited in terms of speed and performance of serial number recognition. Therefore, in this paper, we propose a character extraction method based on the aspect ratio of banknotes and a character recognition method based on a convolutional neural network (CNN). For character extraction, de-skewing was performed first. Then, the serial number was estimated on the basis of the aspect ratio of the banknote. Further, we designed four types of CNN-based neural networks for character recognition and adopted the most appropriate neural network. Subsequently, we confirmed that the average recognition performance per character for each neural network was 99.85%.