Face recognition software has worked in the past like other
forms of 'biometric' identification, such as voice, iris or fingerprint
identification: a computer analyzes a given photo or other biometric fragment
and looks for a very specific set of markers. Comparing facial aspects in this
way is conceptually similar to comparing fingerprint lines, though much more
complicated. If the program finds a critical similarity threshold between the
sample and sample pattern, a match will be declared - so simple. This worked
well enough for relatively simple jobs, such as finding out where faces are in
a photo, but can this face really be identified as matching another photo of
the same person? It turned out to be much more difficult.
Face recognition software and algorithms
There are many methods to help you identify your face.
Gradients
One of them is essentially replacing the image with a
version that emphasizes the most important details of facial identification.
For gradients, this involves replacing each pixel with a
representation of its brightness compared to the surrounding pixels.
This relative measure of pixel brightness makes it much
easier to recognize the same face as the same face in many different lighting
situations.
Relative lighting attributes appear to be true between
shots, while objective lighting is much more variable - but even with this and
other techniques, very different lighting conditions are still a problem for
many modern face recognition systems. (It should also be noted that they impede
human face evaluation.)
Performance
Another approach concerns the so-called
"projection" of a 2D photo onto a 3D model, such as a cylinder.
Wrapping a face around the third dimension can often
reveal forms of symmetry and distinctive features that are much harder to find
on a flat and static image.
After completing the entire image preparation, the system
finally "codes" the face or collapses its most distinctive features
and patterns into a smaller, simplified file that only exists for
cross-checking with other encoded faces.
So, when a picture of Leonardo DiCaprio was shown, this
type of system would first warp and analyze the picture in various ways to
generate an encoded version, and then compare this encoded face with a
collection of encoded faces in a file.
These saved faces are the basis of a comparison to find
face matches and these saved files can be pre-associated with information such
as names and addresses.
