visitor (0 QPoints)
  • FR
  • EN
  • NL
  • DE
  • ES
315 experts, 1193 registered users, 1659 questions already answered
European Experts Exchange, the very best site for high-quality IT solutions

New Improved Search!

 


05/10/2011 1h30 : Steve Jobs is dead, the father of Apple ][ is gone, we are all orphaned.

Languages :: Delphi :: Select all faces


By: Comboy U.S.A.  Date: 06/09/2003 00:00:00  English  Points: 100 Status: Answered
Quality : Excellent
Hi experts,
a program which can find all human faces in an image. i see something similar in "microsoft windows picture and fax viewer" of windows longhorn m4, i think "poser 4' has got this stuff, too. so how can i recognize faces? i wanna anlyze them does learning AI help me about this?
thanks in advance,

Comboy
By: Workshop_Alex Date: 06/09/2003 23:49:00 English  Type : Comment
Compare it with shooting a rocket into space. You'll have to learn about the techniques to do it, the requirements, the mathematics and how to combine it. Then find and use the right tools to start building and finally be ready to shoot and hope it won't blow in your face...
But as time progresses, there will be socalled make-your-own-rocket pachakes for those with less experience. Stuff prepackaged and all you have to do is assemble it. (Then shoot the rocket and still hope it doesn't blow.)

For finding human faces in an image, you'll need something similar. You need to know an algorithm that is capable of detecting certain shapes in an image. This algorithm should be smart enough to see the difference between a face and a doorknob. You can compare it with OCR software, which is able to recognise letters. This is a bit more difficult though since OCR can simply work with black and white images while face recognition uses images with at least many different shades of grey.
But as with the build-your-own-rocket package you might use some OCX library in the near future that will do the recognition for you. But it will probably take a while before it's in the market. Until then, rocket-science...
By: VGR Date: 06/09/2003 23:53:00 English  Type : Answer
look for "biometrics" or "physiognomony" (one more invention of ours)
By: Comboy Date: 07/09/2003 00:23:00 English  Type : Comment
To workshop_alex > thanks for disappointing me :( take a look at this shooting rocket
<A HREF="http://www.comboy.150m.com/face.zip">http://www.comboy.150m.com/face.zip</a>
To VGR > Hmm, sounds great, but give me a few days to find some good articles about them.
By: Comboy Date: 07/09/2003 00:45:00 English  Type : Comment
Sorry, the zip file isn't there, try these instead:
<A HREF="http://www.comboy.150m.com/A.jpg">http://www.comboy.150m.com/A.jpg</a>
<A HREF="http://www.comboy.150m.com/B.jpg">http://www.comboy.150m.com/B.jpg</a>
By: Workshop_Alex Date: 07/09/2003 22:45:00 English  Type : Assist
You're welcome. And yes, I know it's a bit disappointing. Like all new techniques, useful resources are quite scarse in the beginning. So it will be a lot of trial-and-error at first, spending weeks, months even finding useful pieces of code that are probably written in another language. (C++) So you have to convert a lot, write your own algorithms and really try to understand how these techniques work.
There are a few face recognition products already available, I think. But they are mostly integrated in experimental security solutions and thus not really available. You mentioned "microsoft windows picture and fax viewer" and I would advise you to see if you can import some of the modules of this thing as type libraries in your project. If you can, you'll have to find out how to use these interfaces but it will give you access to face recognition in that case.

However, counting faces in a picture can be quite difficult. People can wear (sun) glasses, eye-patches, beards, long hair, ugly scars, piercings or just have their faces painted like clowns for fun. And what if the picture has a movie poster in the background? All this complexity will make the computer less accurate unless you keep all these things in account. If you put two grapes and a banana in a certain position on a table, the computer must be smart enough to see this is not a face. But a bearded guy wearing sunglasses and a baseball cap should be recognised by the system... And no, I'm not going to shave just so your face counter can count my face. :-)

As VGR said, you'll have to search for the right keywords. So the first step is finding which keywords will give you the most results. Also try to contact other researchers who are researching these face recognition techniques. It's a lot of work but that's basically because no one is working to create the simple construction-kits yet.
By: Comboy Date: 07/09/2003 23:26:00 English  Type : Comment
Thank you workshop_alex,
I think if I know more about these thing, it would make it closer:
- how to simplize a picture, i mean find edges and convert it into lines?
Can you help me with this, i guess the first step is to make it greyscale, then lines, then smooth lines, then verify separate zones, then give each zone a color. Umm, am I right?
I'm studying AI on my own, and I want to check kids painting from a kindergarten in our city. I'm crazy about making it possible. OK, i don't want to make a 100% dependable code.
thanks for your advice. more ideas makes it even better :) don't worry about points

Comboy

{computer is the world chess champion}

Do register to be able to answer

EContact
browser fav
page generated in 351.406810 milliseconds

Why Google AdSense ads ?

compteur
 Ranking-Hits PageRank for this page