Everyone - even an identical twin - has unique iris patterns. While fingerprint scans can change over time and be obscured by moist, greasy hands, irises are stable and have false positive rates of less than one in a million.
However, current iris recognition methodologies rely on expensive cameras and near-infrared led lighting systems, costing between $200 to $300. Our objective was to create an effective and cost-efficient technology that harnesses the power of smartphones and functions in the visible light spectrum. To attain high resolution iris images, we built a cheap lens attachment, at the cost of a few pennies with parts from a used disposable camera. Using OpenCV, we then developed a sophisticated computer vision pipeline, segmenting irises and pupils with Gaussian smoothing, Canny Edge detection, and Hough Circle Transforms. Finally, we matched iris patterns, texture, and hue with histogram analysis and SIFT features.
The resulting technology not only can be widely applied for personal use (for example, home security and automatic log-in and payment), but also lends application to issues such as border and airport security.