Facial Recognition Technology
Facial Recognition Technology is a type of biometrics that uses mathematical algorithms to reduce a facial image to numeric data that can be compared to stored numeric data representations of known faces. Findings are represented as a percentage match against the knowns. All facial recognition technologies use the same basic sequence of actions: Obtain the image, Locate the face in the image, Generate a template from the image using the facial recognition software, Compare to stored templates in database, Determine positive matches, and Declare the matches. This can be done manually or automatically in real time or asynchronously. There are 3 methods used: Holistic, Feature-based, and Hybrid.
Methods of Facial Recognition. Holistic methods use the eigenface technique of PCA(principle component analysis) to reduce a facial image to a little more than 100 different characteristics that are assigned mathematical weights using a unified face created by a training database. Known image's features are given weights against the training database's unified facial representation(or eigenface). Unknown's features are then assigned weights and the weights are compared to the known faces. There are several other holistic methods, but they all represent variations on eigenface using somewhat different mathematical methods, e.g. Fisherface using FLD. Holistic methods allow real-time facial recognition as they can dynamically identify all the faces in a crowd, normalize them for comparison, reduce them to a mathematical representation, and compare them to a database of known faces. Feature-based Methods use the distances between features expressed as a ratio to a central point. These were the original methods of facial recognition. Originally done manually and asynchronously, several variations in this category have been automated. Hybrid Methods combine Holistic and Feature-based methods to improve yield. Most proprietary systems currently are Hybrids.
Research and Application Facial recognition has been tried in the field(Tampa,England,Logan Airport,etc) with mixed results since the breakthrough discovery of PCA and the eigenface technique in the early 90s. The FERET database of known faces was developed with DARPA funding in the 90s to allow benchmarking for the comparison/evaluation of new systems. 3 FRVTs(Facial Recognition Vendor Tests) and a FRGC(Facial Recognition Grand Challenge) have occurred since then with very active research underway. There are many vendors and types of proprietary systems available and under development. It is a nearly ideal biometric for law enforcement and counter-terrorism as it can be done covertly and in real time. It has a role in health care for biometric login authentication.
Thomas Carr
