Privacy Enhancing Revocable Biometric Identities (PERBI), 2016-19 –AwardedbyBoard of Research in Nuclear Sciences(BRNS), Department of Atomic Energy, Govt of India(Funding Rs. 28.8 lacks). (Status: On-going)
I contributed to the above live projects under the supervision of Dr. Pritee Khanna(Principal Investigator) at Indian Institute of Information technology, and built a Multimodal Biometric System with Recocable Capabilities, and Implemented a dorsal hand vein authentication system from scratch.
Major Contribution:
-
During Processing Dorsal Hand Image, developed a heuristic blurring method to remove hair from the image.
-
Extracted clear vein pattern from the dorsal Image by Implementing Maximum Curvature Method.
-
Implemented a Cancellable capability to address the Stolen token(user information) scenario.
-
Reduce the dimension of data up to 75% and achieved 94% multi-class classification accuracy•
In this project, I had to build a biometric authentication system, and need to perform authentication based on a person's vein pattern. Since veins are present inside the body, in most cases, not visible to the naked eye so this system was more secure than others.
My projects mainly consisted of five steps which are as follows,
- Data Collection, Dorsal Hand Image Acquisition.
- Data processing.
- Feature extraction.
- Feature protection.
- Classification.
About Data
I used 850 dorsal hand images, taken from a special device that use Infrared light to capture viens data. 850 Biometric Data = 85 person, 10 samples each.
How data capture device work: Unlikely to oxygenated hemoglobin, deoxidized hemoglobin absorbs light at a wavelength of about 760nm, which is in the range of the near-infrared (NIR) band. Therefore, when the dorsal hand is illuminated with near-infrared light, deoxidized hemoglobin absorbs the light and appears as a black pattern in the dorsal image.
About Pre-Processing Data
About Feature Extraction
About Feature Protection
About Classification