6.7 Attrasoft Facial Recognition Classifications System: A Demonstration


6.7 Attrasoft Facial Recognition Classifications System: A Demonstration

Attrasoft is a neural network company selling various systems in the areas of image and facial recognition. For this demonstration, we are using the facial recognition systems and a database of images of known criminals. The Attrasoft system is first trained to recognize a perpetrator and then search for a specific photograph(s) or sketch(s) of the matching suspect. Figure 6.1 shows a photograph of a suspect the police are searching for.

click to expand
Figure 6.1: This is the suspect the police are searching for.

The image database of known criminals is passed through the system; this database can contain an unlimited number of suspect photographs. Attrasoft can search this database at a rate of 10 photographs per second. The first step is to have the Attrasoft system trained to recognize the face of the suspect. This is the ImageFinder interface (see Figure 6.2).

click to expand
Figure 6.2: Attrasoft ImageFinder during training.

Next, the user clicks on the Train button and waits for one second until in the status text area the message "Training End!" appears. The user can modify the setting parameters, like blurring, sensitivity, external weight cut, image type, segment size, etc. Once training is complete, the system can be directed to go out and look for images that match the training sample, with the output having an integer, representing a similarity value. The higher the score between the training image(s) and the retrieved images, the better the match. Figure 6.3 shows high matching similarity translation symmetry and is, in fact, a photograph of the suspect.


Figure 6.3: System recognized the suspect wearing a hat.

However, the system also matched the following photograph based on the original training image (see Figure 6.4).


Figure 6.4: System recognized suspect with a beard.

All three images are of the same person. The facial recognition system was able to make a match despite the fact the other photographs had the suspect wearing a hat and a beard.

Facial recognition software works by measuring a face according to its peaks and valleys—such as the tip of the nose, the depth of the eye sockets—which are known as nodal points. A human face has 80 nodal points; however, facial recognition software may require only 14 to 22 to make a match, concentrating on the inner region of the face, which runs from temple to temple and just over the lip, called the "golden triangle." This is the most stable area because if an individual grows a beard, puts on glasses, gains weight or ages, this region tends not to be affected. The relative positions of these points are converted into a long string of numbers, known as a face print.

This type of facial recognition technology can be incorporated into the Transportation Safety Administration's planned computer-assisted passenger prescreening system (CAPPS II), enabling it to recognize potential hijackers and terrorists from existing databases of photographs. Currently, most facial recognition systems are in use by casinos, with more than 100 across the United States using them in their daily operations.




Investigative Data Mining for Security and Criminal Detection
Investigative Data Mining for Security and Criminal Detection
ISBN: 0750676132
EAN: 2147483647
Year: 2005
Pages: 232
Authors: Jesus Mena

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net