A Camera Phone Based Currency Reader for the Visually Impaired
Xu Liu · 2008 · Proceedings of the 10th International ACM SIGACCESS Conference on Computers and Accessibility (Assets '08) · doi:10.1145/1414471.1414551
Summary
This paper presents a camera phone-based system for identifying U.S. paper currency denominations for people who are blind or visually impaired. The work addresses a specific accessibility gap: unlike currencies in many other countries, U.S. paper bills are identical in size and texture across all denominations, making them impossible to distinguish by touch. The blind community had recently won a discrimination lawsuit against the Department of the Treasury (May 2008), but physical currency redesign would be a prolonged and expensive process. Existing dedicated currency readers like the Kurzweil reader were bulky and expensive. The author proposes leveraging the ubiquitous camera phone as an inexpensive, portable alternative. A key design insight is that visually impaired users cannot be expected to capture high-quality photographs, so the system processes the live video stream in real time at approximately 10 frames per second, providing instant feedback as the camera approaches the bill rather than requiring a single well-framed snapshot.
Key findings
The system uses a multi-stage pipeline: background subtraction via binarisation (exploiting the white border of U.S. bills), noise removal through breadth-first search from the image centre, perspective correction to normalise the detected bill region to a 4:1 aspect ratio rectangle, and classification using an AdaBoost framework. The classifier uses 10 weak classifiers, each based on 32 random pixel pairs whose relative brightness remains stable across lighting conditions — requiring only 320 pairwise pixel comparisons per frame, making it computationally efficient for mobile hardware. The system was trained on 1,000 sample images per bill side plus 10,000 non-currency images, achieving a false positive rate below 1 in 10,000. It recognises , , , , , and bills from either side, with results communicated via speech output or vibration. The framework runs on Symbian and Windows Mobile platforms and is extensible to new bill designs. The author notes that software distribution to end users remains a challenge, as visually impaired people may lack the ability to independently download and install such applications.
Relevance
This paper represents an early example of using mobile phone cameras and machine learning for real-time assistive object recognition — an approach that has since become mainstream through apps like Seeing AI and Be My Eyes. The design decision to process live video rather than requiring a single snapshot was prescient; it acknowledged that accessibility must accommodate how users actually interact with technology, not how designers assume they will. The underlying legal context (the 2008 Treasury lawsuit) highlights how currency design can be an accessibility barrier, an issue that remains relevant globally. For practitioners, the paper demonstrates that effective assistive technology can be built on commodity hardware without specialised sensors, and that the real bottleneck is often distribution and deployment rather than technical capability. The framework's extensibility to other object recognition tasks foreshadowed the broader category of visual assistance tools that have become essential for independent living.
Tags: visual impairment · computer vision · assistive technology · mobile accessibility · object recognition · financial accessibility · machine learning · independent living