SoylentNews is people
SoylentNews
from the combine-with-automated-facial-recognition dept.
KOMO TV (Seattle) is carrying a story about unsolved "Cold Case" murders in Tacoma that occurred in 1986.
TACOMA, Wash. - Using cutting-edge technology not available until now, investigators have released composite sketches of two men suspected of abducting and killing two young Tacoma girls in 1986.
Police say they are determined to solve the two horrific murder cases, which have gone cold after three decades - and they are hopeful the new technology will help lead them to the killers.
There were no witnesses. But DNA samples were found. So how were the sketches made?
The "composite sketches" were generated by a computer based on a process called DNA Phenotyping which is the prediction of physical appearance, using information extracted from DNA which accurately predicts genetic ancestry, eye color, hair color, skin color, freckling, and face shape in individuals from any ethnic background, even individuals with mixed ancestry.
"These are composites much like a witness giving a description and a computer program making a sketch based on known appearance factors," Loretta Cool of the Tacoma police said in a prepared statement. "These composites will not be exact but the outcome is a visual reference that may look similar to what the suspects looked like in 1986."
The process was developed by Parabon Nanolabs and the process is explained on their web site.
How close are the predictions?
Parabon's website has some examples generated from DNA contributed by known volunteers. You can compare the sketches with photos of the volunteers and judge for yourself. Personally, I think Yolanda McClary's actual IMDB photo is virtually a dead ringer for the computer prediction.
(Score: 2) by Aiwendil on Friday April 08 2016, @11:51AM
Ok, I'm bad at recognizing faces to the point where I have troubles recognizing my own mother in pictures of her face but I can often recognize if it is the same person in two images placed next to each other.
However, the example-pictures are what I would consider "not even close", this makes me wonder just how much this relies on the brains habit to fudge data in order for them to seem similar.
For instance the eyecolour, skincolour, skeletal features of cheekbones and forehead, and the dimple at the chin are all completly off. The only thing I would say this got right was the structural features of the eyes-nose-mouth and roughly if they where black/white/east-asian.
(Score: 0) by Anonymous Coward on Friday April 08 2016, @01:10PM
The right way to check it is to select a bunch of sufficiently similar looking volunteers, generate "DNA images" from them, and then give the real and computer generated images to test subjects who have to find out which generated face belongs to which real face. The accuracy of the method would then be reflected in the fraction of correct pairings, compared to the expected fraction for pure random pairings.
(Score: 2) by frojack on Saturday April 09 2016, @03:21AM
However, the example-pictures are what I would consider "not even close",
I thought they did look amazingly close, give that they were made from a cheek swab, especially if you have seen these people on TV previously. Like I mentioned, Yolanda's imdb photo is a very close match to the generated image. Kate? Not much.
They are certainly good enough to point you in the right direction when someone in the neighborhood thinks it kinda sorta looks like Joe, but totally different than Bob.
Remember they still have the DNA for an actual match.
No, you are mistaken. I've always had this sig.
(Score: 2) by Aiwendil on Saturday April 09 2016, @06:28AM
Our differences in opinion regarding if they are close or not is why I wonder just how much it relies on the brain to make a bunch of leaps.
I assume you have normal or better ability to recognize faces which [if true] would mean you identify people on different features than I do.
If you have the time I'm curious about if you still conside them similar when viewed upside down (or section-by-section when split into 1x4 or 3x3 sectiond), just to delay lots of the instinctive habits.