Facial Recognition Algorithm
Multivariate State Estimation Technique (MSET) for Pattern Recognition and Multi-Class Image Classification
Available for Licensing
US Utility Patent: US 7917540
At A Glance
- Object recognition technology for video processing even with substantial condition variability.
- Reliably identifies faces from data sets with large variations in illumination.
- Resolution requirements are minimal – commercially viable computing requirements.
- Wide-ranging applications, from security and video surveillance to cancer detection.
- Possible end-users: banks, airports, Dept. of Homeland Security, retail stores, and more.
The realization of object recognition software has the potential for wide-ranging impact, from cancer detection to video surveillance. Development of such technology, however, has been stymied by the intractable problem of condition variability. For example, facial recognition technology is typically unable to match an individual with eyeglasses to a reference picture of the same individual without eyeglasses. Other variables, such as make-up, wisps of hair, camera angle and even light conditions further complicate the process.
Researchers at Colorado State University have made significant technical advances in this field. By embracing, rather than compensating for, condition variability, these researchers have developed novel technology capable of identifying individuals even under dramatically different lighting conditions. Their set-to-set algorithms have been proven using the largest data sets available.
Importantly, these researchers have also been able to reduce the video resolution required for reliable object recognition. Incredibly, reliable facial recognition was achieved with images consisting of a mere 25 pixels—a resolution so low that the faces are unrecognizable to human eyes. The minimization of the resolution requirement has important ramifications to the commercial viability of facial recognition technology, as large sets of data (e.g. continuous video monitoring) will not require unreasonably large computer memory or processing power with such low resolution.
This technology has applications to facial recognition in an uncontrolled environment (i.e. not an artificial studio) , video surveillance, automated event detection, and even cancer detection. Possible end-users include banks, airports, Department of Homeland Security, professional sporting leagues, retail stores and more.
- Very fast algorithm
- Natural for representing nonlinear spaces
- Can be implemented in parallel
Last updated: May 2020