VeinDeep: Smartphone unlock using vein patterns
Henry Zhong, Salil S. Kanhere, Chun Tung Chou · 2017
This paper presents VeinDeep, a system for using vein patterns to secure smartphones from opportunistic access, e.g. a device left unattended. VeinDeep takes advantage of infrared depth sensors, which at the time of writing have recently started to appear in smartphones for 3D indoor mapping and localisation. We find these sensors can be re-purposed to capture images of the unique vein patterns on the back of each person's hand. We simulate a depth sensor equipped smartphone by developing VeinDeep on a low power Compute Stick. We use Kinect V2 depth sensor to collect 240 recordings from 40 hands belonging to 20 test subjects. Then use this data to compare VeinDeep to one older but popular vein pattern recognition algorithm which uses Hausdorff distance and one recently developed algorithm which uses Kernel distance. We achieve a precision of 0.98, compared to 0.9 for Kernel distance and 0.5 for Hausdorff distance when recall is approximately at 0.83. In addition VeinDeep does all this while taking an average of 6 MiB of memory and 466 milliseconds per comparison. This is an average of 1/6 run time, 2/3 the memory of Hausdorff distance and 1/3 the run time, 1/2 the memory of Kernel distance.