Slashdot Mirror
The Face Detector
Roland Piquepaille writes "Almost all human faces have common characteristics, such as two eyes and one mouth. Still, some people, affected by face blindness, cannot recognize one face from another one. So it's understandable that face recognition is a major challenge for computer vision systems. In "Facing facts in computer recognition,", the Pittsburgh Post-Gazette reports that a team from Carnegie Mellon University's Robotics Institute has developed a very accurate software to find faces within images. By analyzing only 768 pixels, the system can detect 93 percent of the faces in a set of images while falsely identifying four objects as faces. The Face Detector Demo is available online and you can submit an image for analysis and receive the results by e-mail. The technology will be used for security purposes, but also by digital photography companies who want to automatically reduce "red eye" effects. You'll find more details and references in this overview."
Yes, yes it does. This one of the big problems with the software, is that some things look like faces and really aren't. A human can tell because we've got a lot more training on different data sets. After seeing some of the demos of this stuff, either they really jacked up the accuracy in the last few weeks, or it was under more controlled settings. Off a picture from a new york street it could only pick up about 60% of the faces and had a decent amount of false positives.
Also, for those who won't read the article, this is just about finding the faces, not recognizing them. This is a prerequisite toward ubiquitous facial recognition.
My Slashdot account is old enough to drink...
Here
I think this technology just recognizes faces from backgrounds, it *does not* appear to uniquely identify faces (a la fingerprints).
Others have tried that, and we all know how monumentally insufficient it has been thus far as a legitimate security tool, in terms of missed matches and a high false-positive to actual positive ratio.
Xentax
You shouldn't verb words.
You're crazy.
Now that we have that out of the way, "face blindness" is a mental condition where the part of the brain that normally remembers what a face looks like does not work properly (somewhat like a learning disability). It has nothing to do with the person's sight, but with their memory.
It's rare, and quite odd, but it's recognized as an actual disorder.
I googled the topic and found some really amazing information. This site covers this topic greatly. I was amazed at this illness. I am no were near a great person at remembering faces but I think it's amazing that if I walked up to you and said hello, then walked away for 10 mins and came back you wouldn't recognize me. It's like Finding Nemo all over again. :)
http://jayceecorder.blogspot.com
While I agree with you about the general "culture of euphemism," as you put it-- I don't think this is one of those times. Face Blindness is not referring to people like you and me who are just lousy at remembering who we met, but rather people with profound neurological disorders who *literally* cannot tell a face from something vaguely facelike, like a vase or a particular arrangement of shadows. This goes far beyond not remembering the guy you met at a convention a year ago-- but rather not even being able to tell the difference between his face and the PDA he was holding.
For a quick read on it, check out The Man Who Mistook His Wife for a Hat. The things that happen to the poor people in this book as a result of disease, physical damage to the brain, or conditions they were born with are bizarre but definitely interesting.
I am no expert in this technology, but I am somewhat knowledgable about it, let me explain something.
You won't understand how hard is it to actually pull off something like face recognition until you yourself actually sit down and try it, only to realize that the problem is much more complex to solve when it has to be so all-encompasing.
The first step to face recognition is to recognize where the face is. The result of this process are quadrilaterals that carve out the face so that when you crop, you are left with exactly the face (frontal, or profile view or other).
A common technique used to do that is to locate the eyes. Most faces (heck, even those with veils on them for relegious reasons!) will contain eyes. Then, when detecting where the face is, you are only left with not having covered people who are wearing sunglasses (which are much easier to detect).
After you have located the eyes, you gauge by their proportions the approximate proportions of the face. Then, you apply an iterative technique (varies in principle, typically based on differential calculus combined with numerical methods of approximation) to locate the bounds of the face so you can eventually crop it to know WHERE THE FACE IS.
"Obviously", the iterative technique has to be able to detect false positives via a threshold set that will rule out the non-face. However, once you have located the eyes with certain reliability, the overall chance that you have come across a face is pretty solid.
The problem is complicated as it is already as you can see!!
Only after FINDING the face, you can start MATCHING the face. At that point you are facing a number of problems that the imagination of most /.-ers can conceive of... Bierds, smiles, teeth-showing, frowns, skin tone changes and the most popular by all scientist: plastic surgery....
A common approach to the actual face matching is a technique of the so-called eigenfaces, whereby you compute a "common" face of the pool and then you can navigate down the specialization of characteristics (e.g. bigger, bigger, bigger nostrils) as you drill down, narrowing down the pool of possible faces.
There is nothing that takes away from how much state-of-the-art CMU's research is. It would be like saying "why is someone dealing with virtual memory management of an operating system if by now, we already have user applications for the OS". Do you see the flaw in such thinking?
The science behind is a lot of mathematics, so dear parent, please don't be ignorant of this type of work just because you don't understand its complexities...
It's the latter. Except it's probably not really "random" - there are, of course, reasons that those 4 gave a false +. It's just that we don't know why (unless you consider ``the algorithm is not sufficiently advanced'' to be an explanation, which I don't).
On another note, it is a very stupid writeup which gives a ratio for correctness and an absolute value for the number of type I error, without even giving the size of the dataset. For shame...
The difference, is something I've had to deal with at various jobs. Although I can after seeing a person for quite some time (several hours or meeting/talking with them 5-6 times for an hour or more) I can keep their face and name in memory, I cannot do this on a short term basis. Example. I've worked in a grocery store and a K-Mart. People will often stop you to ask where something is, and, if the item is out, you have to go check in the back for them. When people would ask me for something, and I'd have to go to the back, I'd have a problem finding them if they moved from the time I left to the time I came back. Half the time, I would be lucky enough that either the person didnt move, was waiting for me and spoke up, or was wearing some ungodly horrible color of clothing that I could instantly identify and find. The other half of the time, the person was wearing something dull and wandered off a little way and I would have no chance to recognize them unless they came back to me. Really, its a pain in the ass to be that way, as your always wondering if your going to find the right person when you get back from something. But its nothing you cant live with... I'd imagine it'd be horrible for it to get worse and not recognize family...
~~ Please keep your arms, legs, and outright stupidity inside the ride at all times. Thank You ~~
The dataset they are probably using is the CMU+MIT hard face detection dataset available here (link to 30 meg tar file at bottom of page): http://vasc.ri.cmu.edu/idb/html/face/frontal_image s/
This dataset is made of images that contain static, contain faces at unique angles, contain complex scenery which would mess up the way most face detection algorithms work, and more "hard" cases. It's the dataset that all face detection programs are tested against. If I remember correctly most other face detection programs only achieve approixmatly 80-85% detection rate when they only get 4 false positives in the results, so this algorithm is definatly an improvement.
Faces are moderately recognizable for me, but no more so than other objects, like rocks or cars. See this page, where someone else with this problem has a demonstration page.
Prosopagnosia is rare. Only about thirty people in the US have been formally diagnosed. I have; there's a researcher at U.C. Berkeley who ran me through the tests. There's a specific section of the brain that does face recognition, and it goes active when looking at a face. This can be detected with a functional MRI scan, which takes hours and involves looking at pictures while inside an MRI machine. For people with prosopagnosia, that doesn't happen.
It's a social handicap. It's most annoying in medium-sized groups. In big groups, you're not expected to know everyone; in small groups, cues other than faces are sufficient. It's subtle. One of the most subtle effects is that recognition takes well under a second for people who can recognize faces, but may take two or three seconds if you have to do it by other means. This breaks some implicit social cues.
Recognizing people works about equally well from the front, side, and back for me. Voice and walk are more helpful than seeing the face.
Practice doesn't help. The non-face recognition skills can be improved, but that's a workaround. Real face recognition, the kind that makes reading People magazine meaningful, is totally out of reach.
"Falling in love" doesn't work, either. Sex, yes; friendship, yes, love, no. That's tied to face recognition. As a friend of mine puts it, "there's no click".
It hasn't interfered with professional success for me; I have an advanced degree from a big-name school and I'm a multimillionaire. But it leads to a strange life.
http://vasc.ri.cmu.edu/demos/faceindex/05062004/us ers/2250.html
The hard part is figuring out the little details that often get inexplicably omitted from journal papers. What are the particulars of the dataset? How are the training images preprocessed? What is the arbitration strategy for overlapping detections? These are the types of details that seperate the output quality of systems that use identical algorithms. In many cases, the researchers are happy to answer questions via email, unless they have plans to spin the research off into a private company.