Current Approaches

Because of the difficulty of the face detection problems, there have been a variety of techniques created to try to provide a solution. These techniques range from simple spatial filtering and thresholding based on flesh-tones, to complex probability models based on Gaussian distributions and Eigenspaces.

Several techniques are discussed at a Face Detection Home Page set up by Dr. Robert Frischholz. He states that the problem of facial recognition begins with facial detection. Once faces can be reliably detected "in the field", recognition can then take place to see who it is.

One technique that is discussed in detail in a paper is a flesh-tone detection in various lighting condiditions. Results from this paper look promising, but unfortunately, the specific details of the algorithm are left out.

The advantage to using infrared illumination to these other techniques it that it exploits a well-known eye reflex that is present in most of the human population. It performs equally well on people of different ethnic backgrounds, as well as size and shape, which is a pitfall of quite a few other approaches where consistency of features is an obstacle. This technique also scales extremely well.

[Introduction| Current Approaches | Hardware | Software | Problems | Future Research | References]