Project
| # | Title | Team Members | TA | Documents | Sponsor |
|---|---|---|---|---|---|
| 1 | Parallelizable Algorithm for Hyperspectral Biometrics |
Akshay Malik Christopher Baker Timothee Bouhour |
appendix0.pdf design_document0.pdf final_paper0.pdf other0.pdf other0.pdf presentation0.pdf proposal0.pdf |
||
| In the modern world unique person identification has become an increasing challenge, central to strategies in combating terrorism and crime to provide global security. Recent research has shown that hyperspectral imaging provides new and improved biometric data, which can be leveraged to meet this challenge by examining features in different spectral bands. Our product uses hyperspectral imaging to correctly identify a given person from a database. This product will: 1) Use as an input existing raw data from hyperspectral sensors (from external sources) which would have, for each point in an image, both coordinates as well as intensity for a given range of hyperspectral bands. 2) Determine which points of the image to be used for analysis and comparison (such as: skin between the eyes, cheeks, forehead, and nose bridge). 3) Develop a comparison algorithm and use the parallel processing capabilities of a GPU to run this algorithm with the band data for these points and the data in the database in the most efficient way. 4) Output the closest match in the database and how similar it is to the target as a percentage. 5) Remotely activate a door locking mechanism if a match is found within a given accuracy. We will preload our database with data gathered from a variety of test subjects, then try to match the original images as well as additional images of the same subjects to the database. Our stretch goal would be, if time permits, to enhance the algorithm to account for differences in facial expressions, makeup, lighting source, head orientation, etc. |
|||||