Face Recognition has been a highly active research area for
many years for its wide potential applications, such as law enforcement,
security access, and man-machine interaction. The existing techniques have
reached maturity in their capability but cannot still fully address practical
challenges including the problem of recognition under varying facial
expressions. Another commercially important challenge is handling high
dimensionality data for large face databases, i.e., dimensionality reduction
without much compromise on accuracy of recognition.
Researchers at Arizona State University have developed a
novel technique that can achieve super-compact, space efficient, highly
compressed, intelligent feature extraction with varying expressions, that
enables to compactly pack the information pertaining to facial expressions (from
many images of an subject) into only two feature images, with negligible
information loss for recognition. This includes critically down-sampled face
images or its ultra-low dimensional random projection which can contain as low
as, say just 25 data points, for each subject in a fairly large face database
(with thousands of images and hundreds of subjects). Typically the existing
techniques need far higher dimensions, at the least of hundreds of data points
to achieve good classification results.
Potential Applications
- Face Recognition systems
- Compact imaging solutions
- Digital storage media
Benefits and Advantages
- Super compact, space efficient, and highly compressed
face-representation and feature extraction
- Ultra-low operating dimensionality for recognition (at
least 4 times less than existing techniques)
- Integration with Compressed Sensing architecture:
Compressed Sensing technique is a recent innovation which offers many
advantages over conventional imaging.
- Object recognition, ensemble data compression
- Down sampled up to just 25 data points per person,
compared to at least 100s of data points with existing techniques.
Download original PDF
