Facial Recognition Software

There is always a dispute between how much casinos in Vegas use facial recognition software and how good their systems are.

Lately the whole field got stood on its head. Literally. Computers are so much better at the task if the facial images are presented to the computer upside down.

Hey, they work great. In the TV show Las Vegas
they could find somebody is 5 sec or less and
that was 10 years ago. It wasn't fake either,
it looked really real.

Quote: Fleastiff

There is always a dispute between how much casinos in Vegas use facial recognition software and how good their systems are.

Lately the whole field got stood on its head. Literally. Computers are so much better at the task if the facial images are presented to the computer upside down.

Military black projects are sometimes as much as 10 years out before becoming known to the public. Right now I wouldn't be surprised if there is big cash allowance for improved facial recognition.

The ability to quickly sort through crowds and ID people would be pretty high on the list.

Quote: rxwine

The ability to quickly sort through crowds and ID people would be pretty high on the list.

It is. The UK has more cameras than any other nation but very few crimes are really solved using cameras, manpower is tied down and often misinterpretations of behavior takes place. London and surrounding areas can trace vehicles "live" but its hard to trace pedestrians even if they stand out by running and even more difficult to recognize pedestrians through facial recognition programs.

Computers often have trouble compiling a face from bad cameras and bad angles but by flipping crowd images upside down facial scanning is highly improved. And smaller and smaler subsets of candidates get presented as possibilities.

A deep convolutional artificial neural network can "fill in" areas of faces covered by scarfs, wigs, or blocked from the camera by other persons or objects or raindrops. The multi layerd neural network than processes the resulting enhanced image in what is almost "real time".

Annoucement here

Multi View Face Detection Using Deep Convoluted Neural Networks.

No need for any annotation or markup, no need for segmentation or feature analysis or SVM calculations.

The major point to note is that this amazing breakthrough in speed and accuracy of facial recognition was made by feeding a "training set" of images to a neural network that consisted largely of images that contained no faces and no humans, only a very few faces were contained in the "training set". This is of course contrary to the usual training techniques on narrow margin classifiers.

In the early 1990's I worked on an imaging project to detect marijuana plants from aerial photographs. I would imagine that today the government has very good facial recognition systems.

Quote: DRich

In the early 1990's I worked on an imaging project to detect marijuana plants from aerial photographs.

One of the earliest projects was an ANN to detect camoflauged tanks. The device was presented with two sets of training photos: tanks in the open and tanks hidden by foliage. The ANN soon developed 100 percent accuracy... but when some general walked in with new photos, it showed that all the device had learned was to discriminate between photos taken on a clear day and photos taken on an overcast day.

I had a dream knowing who knew all, which they knew, and they tried killing me to protect their secrets, with the means to cover up. This gets scary with more technology where an end game evolves like this.