Facial Recognition is something most people have heard of in the media, whether it be NCIS, or the news. In reality, most of what we thought we knew is false when it comes down to analyzing data in an investigation. You don't run fingerprints through a program and get an instant hit. It is done by hand, point by point, and isn't always accurate. Facial Recognition, on the other hand, has been pursued with success in small test groups. The extent of this success, however, is limited. Out of a few thousand pictures, there are multiple algorithms that have proven themselves useful, but what about the entire population? Recent research has shown that most of these algorithms dramatically dropped in accuracy as more people were added to the test group. It became to difficult for the programs to sort through and pinpoint one person in a larger group of photos. Still, there was one way that seemed promising. Having an algorithm teach itself how to make accurate matches in a very large test group from the beginning seems to be more successful. The programs are, in a sense, "used" to extensive sorting and picking through a large group of faces.
Now, researchers are trying to make these algorithms more accurate by sending multiple pictures of the same person at different ages and angles through in hopes to train the computer how to reason and account for variability in different photographs.
Technology is ever advancing!
Article: http://phys.org/news/2016-06-facial-recognition-algorithms-cope-million.html
No comments:
Post a Comment