AI for Good

Chapter 18BankNote-Net: Open Dataset for Assistive Universal Currency Recognition

—Felipe Oviedo and Saqib Shaikh

Executive Summary

Millions of people around the world have low or no vision. Assistive software applications—including optical character recognition, scene identification, person recognition, and currency recognition—have been developed to help such people engage in day-to-day activities. Recognition of different denominations of banknotes can be accomplished using computer vision models; however, the datasets and models available for this task are limited, both in terms of dataset size and the variety of currencies covered. Here, we collected 24,826 images of banknotes in a variety of assistive settings, spanning 17 currencies and 112 denominations. Using supervised contrastive learning, we developed a machine learning model for universal currency recognition.

This model learned regulation-compliant embeddings of banknote images in a variety of contexts: these can be shared publicly (as a highly compressed vector representation) and can be used to train and test specialized downstream models for any currency, including those not covered by our dataset or for which only a few real images per denomination are available. We deployed a variation of this model for public use in the latest version of Microsoft's Seeing AI application, and we share our encoder model and the embeddings as an open dataset in a BankNote-Net repository.

Why Is This Important?

Visual ...