Overview / Usage

While biometrics have become a vital part of user authentication in technological platforms, they suffer from issues like fabrication, high costs, and vulnerability to physical damage that make them difficult to implement in low-cost scenarios. To meet this need, this project authenticates identities through inexpensive peripheral vein biometrics with the Edison Vein ID (EVID) system.

EVID consists of a low-cost NIR camera and low-power NIR LED array. The array emits ~850 nanometer light, which is absorbed by the finger veins’ deoxygenated blood and detected by the camera. Through the Intel Edison Compute Model, a computer vision algorithm captures a raw image, performs contrast-limited adaptive histogram equalization to increase the veins’ contrast, and reduces image noise through a Bilateral Gaussian Filter that preserves the veins’ edges. Finally, a binary threshold segments the image and extracts the unique vein structure, which is then registered as a biometric template.

A MATLAB normalized cross correlation algorithm computes a matching score [0-1] to compare an input image against a template for authentication. This process was repeated with all registered templates to test EVID’s ability to use input images to identify individuals. A 0.573 match score threshold for the verification and identification processes was determined based on False Acceptance Rates and False Rejection Rates according to Biometric Evaluation Standards.

EVID captured, processed, and stored peripheral vein images from 30+ samples, using these templates to authenticate and identify individuals from over 170 images. Future work includes using multi-modular biometrics to create the most secure identity verification system.

Methodology / Approach

This system consists of a custom near-infrared (NIR) camera, low-power NIR LED array and fingerprint sensor. The LEDs emit ~890 nm IR light, which is absorbed by deoxygenated blood in the veins of an individual’s finger. The camera captures light that passes through tissue around the blood, producing high-quality images of the finger’s veins. A computer vision algorithm then determines regions of interest through centroid analysis, performs a contrast-limited adaptive histogram equalization to increase the veins’ contrast and reduces image noise via a bilateral gaussian filter. Finally, an adaptive gaussian binary segmentation algorithm extracts the unique vein structure, which is then registered as a biometric template in the database. This method is coupled with a fingerprint sensor that captures minutiae features of an individual’s fingerprint through a Raspberry Pi compute module. The system was used to capture, process and store bimodal (fingerprint and finger vein) templates from 50 sample index fingers.

To authenticate an individual against stored biometric templates, the system uses a MATLAB cross-correlation template matching algorithm to compute a matching score between 0 and 1 for the individual’s bimodal input with each registered template. This process was repeated with all registered templates to test the system’s ability to accurately identify registered users against “imposters” not registered in the system. A support vector machine model for user authentication was trained and tested against 450 bimodal inputs from “imposters,” demonstrating a True Rejection Rate of 99.98%.

Technologies Used

Intel Edison, Raspberry Pi, Python OpenCV, MATLAB.