What is Digg?16 Comments
- Thrilltone, on 06/18/2008, -1/+10Sadly, this technology will probably just be used to enable computers to better decipher the code we often have to type in, to prove we are human.
- StevieJanowski, on 04/02/2009, -0/+8I feel like I have seen this in action at the photosynth demo for the TED conference. Awesome technology though
- inajeep, on 06/17/2008, -0/+7"The code, in Python and designed to be run within Matlab, is also available under the Creative Commons Attribution License. "
Very nice, the software is at a %16 success rate but I can imagine it improving over time and changes. I wonder how close this is to facial recognition in terms of method. - Pokho, on 06/17/2008, -0/+6I'm also thinking about facial recognition... what if you had a program run through the web finding pictures of faces and correlating it with names attached to the pictures.. then looking up a persons name could result in all pictures where the persons face is in... imagine they make it work for video as well http://www.delvenetworks.com/ and then with all the crappy surveilance cameras around http://tinyurl.com/4ntp8j things might certainly get interesting... face recognition as such is already a feature in google, http://images.google.com/images?hl=en&q=electronic ...
- whiteguysamurai, on 06/18/2008, -0/+5Computers will eventually have to recognize people before they rise up and kill us, makes it easier.
- mymate, on 06/18/2008, -1/+5surely this would be something google should be interested in doing?
- Grimdotdotdot, on 06/18/2008, -0/+4You what?
- Thrilltone, on 06/18/2008, -0/+4I suspect you of being non-human
http://www.webupon.com/Security/The-Human-Verifica ... - Phearce, on 06/18/2008, -0/+2Thinking about facial recognition, there was an old episode of V.I.P. (I know, I know...) where they fed the highway traffic cam feed to a modified facial recognition program in order to id a specific vehicle (they didn't know the plate).
Fast forward 10 years and it's easy to imagine privacy-invasive mashups like Google+Traffic cam+facial and/or vehicle recognition to track people. - yuanzhoulu, on 06/18/2008, -0/+2no it's not 16%. chance is about 1/1000000000 and this software has made it 16/1000000000.
- mikbor, on 06/18/2008, -0/+2Was that the software microsoft was developing?
- nobeastsofierce, on 06/18/2008, -0/+2haha, was gonna say the exact same thing, what is the random occurance of naming a spot anywhere on the earth, within a 200km radius?
- anothrnbdy, on 06/18/2008, -0/+1I believe so.
- wolferz, on 06/18/2008, -0/+1FTA: "The authors were inspired by recent data in cognitive science that suggests we humans do a basic form of image analysis by recognizing when aspects of the image are similar to something we're already aware of."
In other news... water is wet. More on this astounding discovery after the break. /sarcasm
Actually I'm pretty sure this is the ENTIRE way people identify the objects in a picture. Also pretty sure this is something that qualifies as obvious.
BTW... Why not "teach" them to identify concepts first. Seriously, what they are suggesting is a simple (ok not so simple) comparison of an image and several "templates" in it's database. The computer still isn't identifying the items in the picture. It's just regurgitating what some one else programed it to regurgitate when it succeeds at a certain percentile match.
"Teach" a computer to understand concepts such as "screwdriver" Give it a 3d model of a screw driver to turn over and examine to find a match instead of using a static image that will NEVER line up with the image it's viewing. Then when it can identify those readily, give it 3d models of everything you might use a screwdriver on and program it to "know" that there is a relationship between them. Program it to know that when it found a door jam in the picture and a guy is standing in the door looking at the door jam that it is probable that the item in his hand is a screwdriver... and then program it to "know" it that this confluence of images is what is called "a guy installing a strike plate on a door jam with a screwdriver."
I mean, think about it. The human brain doesn't just rifle through every object you have every laid eyes on when looking at a picture. It looks first at the most readily identifiable object, "runs" a contextual check to see what that object is there for and then looks at the objects around it, identifying them based on the context of the first item. This is why those double-image optical illusions work... and why different people see different thing's at first based on which of the items are more readily recognizable. Of note is that most such optical illusions use faces in some respect because the human brain is hard wired to recognize faces more than anything else.
While comparing against a 3d object would significantly increase the time needed to identify each object, using "contextual checks" would significantly reduce the number of comparisons needed. Only occasionally would you have the computer do a "double take" to identify an object that is out of place... like a guy trying to put in a screw with a salami sandwich... for example. - Aisha417, on 06/18/2008, -0/+1That's flippin' sweet
- ktags, on 06/18/2008, -0/+1kind of rip on photosynth which is the coolest bit of software i've seen recently but maybe thats what the "recent research" is. either way good work, and like psogle said if you haven't seen the TED demo of photosynth, check it out: www.ted.com/index.php/talks/view/id/129. Too cool.
© Digg Inc. 2009
— Content created and posted by Digg users is