Researchers at the Naval Research Laboratory have invented powerful facial recognition software. The technology is available for license to businesses that would develop it into a commercially available product.
Interest in computerized facial recognition grew decades ago as the image library of repeat illegal border crossing offenders grew too large to be searched manually. Researchers began to develop algorithms to find a number of images that were likely matches. Images in the database were compared and ranked, based on a certain distance or similarity measure. A human operator would then complete the search by examining the likely matches. The ranking approach became the standard method for academics and commercial companies. However, the ranking approach is not well suited for watchlist or blacklist surveillance.
Researchers at the Naval Research Laboratory developed an improved approach specifically for watchlist surveillance that replicates human facial recognition processes.
The prototype system, known as cFaces, is capable of autonomously recognizing a person of interest and does not need to search any image database, nor does it need to match or rank images as a precursor to recognition.
This novel system is based on identifying and marking the region in the face that belongs to the person of interest. cFaces builds a dedicated classifier for each person of interest using two large sets of borderline images, one set projecting just inside the decision region and one set just outside of the decision region.
Unlike other facial recognition systems, cFaces preserves individual privacy, storing only the images of known bad actors on the watchlist, and by limiting access to authorized users.
- Fast and reliable identification of persons of interest from a comparison data set of original and constructed images
- Effectively increases the training set for cases where adequate facial image data is not available
- Particularly useful for real-world applications where an image is not guaranteed to be in a database
- Businesses can productize the technology by licensing US Patent 7,684,595 from the Navy
- Data is available and research is ongoing