picture recognition software

[spydaz]

HI CARL. ive been looking into what goes into a Picture and identifing Parts of a picture. ie colors and shapes...

as i remember ER1 robot... it has a database of Pictures and atempts a patern match, pretty good.. extrcting the software is expensive...

but a simular thing would be needed.. as hal finds pictures on the web,, ie wot is a cat, hal popsup picture and saves it automatically for clip art. and databasearchive for paturn recognition later...

Would be nice.. Although im not sure if pictures can be saved into a database... perhaps a binery version, to be reconstructed later{when needed}...

Hmm


BiLL.. I think that Image Matching is the right way to go, as hal detects movement from the camera >> Zone in on face and Snap a quick Jpeg.. for paturn matching... Face recognition is at this stage{FEEZE FRAME} we dont have Video recgognition YET.. I play a video trailer, hal matches it to a MOVIE...


Hmmm. that would be nice YumYum

[Bill DeWitt]

quote:Originally posted by spydaz
I think that Image Matching is the right way to go

I can't see anything else happening on Desktops anytime soon. Real object recognition would require both on-the-fly reference modeling, fuzzy logic shape matching and an internal register of models large enough to swamp most hard drives by itself. The CPU demand would be impossible. That's why many mobile robots only recognize obstructions.

Something like this http://arba3d.com/ running once a second or so, (clearly enough to crash a Windows computer) -PLUS the whole model comparison program -PLUS the program which runs the final robot action...

And this doesn't even approach the problem of recognizing that both a Great Dane and a Pomeranian are members of Object_Dog()

[Art]

Bill D.,

I still have Canoma (from Metacreations) which allowed one to create 3D images from photos or other pictures. One could do buildings, furniture and most straight lined objects. The completed image could then be rotated and viewed from different sides, etc.

I also had a webcam program that could recognize various objects once it was "trained." One simply held an object in front of the camera then once the picture was captured the user would type in the name of the object. Apparently the objects were kept in a file with their associated onject names. If one would then hold up say, a pen or watch, the cam program would do a comparison to objects held in the file until a "match" was found, then the program would announce the name of the object it "sees."

I don't recall the program name at the moment but if you are interested I can probably find it for you.

There are also at least 2 chatbot programs that allow the use of a web cam (3 if you count Talking Desktop).

Robert's proposed modifications sound very promising toward raising the bar of Hal's ability. Very cool!