Abstract:

In this study, multiple recognition techniques were experimentally compared to determine the best approach for automatic recognition of Nigerian paper currency notes. Different samples  across all denominations of the Naira were digitized to form the dataset for all the experiments  carried out in this study. Subsequently, pre-processing and feature extraction techniques  including the histogram of oriented gradient, local binary patterns, speeded up robust features and the  scale invariant feature transform among others were employed in order to attain better results from  the selected pattern classifiers. The best recognition accuracy of 88% was achieved using  transfer learning approach as against the 93% (92 seconds time complexity) for the BoVW strategy is deemed acceptable for the development of an automatic Nigerian currency recognition system.