If you want the polygons to be coincident with image regions then do a Goggle search for "image segmentation." Many packages exist to do this, including some optimised for specific CPUs. However, I doubt that you do want this.
I suspect you want the polygons to be coincident with the faces of 3D polyhedra. This problem is hard for 2D image data. It is unsolved in almost all practical cases. The exceptions are mainly industrial inspection of extruded parts, be they cog wheels or biscuits. (Don't ask. Just take it from me that biscuits are not manufactured the way your mom cooks them.) The best you will get is a publicly available, or commercial, library that supplies components to do this job. You will have to do an awful lot of software development if you want to process 2D images bottom-up. Fortunately, converting 3D images to polyhedra is relatively straight forward and commercial packages exist to do this. 3D images usually come from a "laser range finder" so add this term to your Google search, along with "triangulation" or, conceivably, "tessellation." Even so, your AGI might be swamped with an awful lot of triangular facets. You will probably want to use polyhedral, not curved, objects in your real world scenes and clean up the 3D polyhedra to impose planarity. There is another way. If your real world scene contains objects of a known type then "model-based vision" is appropriate. In this paradigm knowledge of the 3D structure of an object is used to search for it in a 2D image. This works well for planar and curved objects providing they are constrained to appear on a known ground surface, but the problem is effectively unsolved for recognising even known objects in an arbitrary pose. I hope this helps, James Anderson -----Original Message----- From: Ben Goertzel [mailto:[EMAIL PROTECTED] Sent: 28 September 2005 01:05 To: [EMAIL PROTECTED] Listbox. Com Subject: [agi] Image to wireframe model conversion Hi all, This list has been way too quiet lately! I plan to post something on Piagetan learning and AGI and try to get an interesting discussion going, but I'll do that next week because I'm away from home right now. In the meantime, I have a question that someone on this list may be able to answer. In our work connecting Novamente with the AGI-SIM simulation world, we have decided to have Novamente visually perceive *polygons* rather than pixels or objects. (The reason is that giving it objects as percepts is too much cheating and denies the system the ability to learn on its own what an "object" is; whereas, giving it pixels as objects would require us to basically build a visual cortex for Novamente, which would be a lot of work and is something I'd like to avoid for the time being.) This is fine for AGI-SIM in which all objects are explicitly made of polygons, but doesn't do us much good in terms of hooking Novamente up to actual real-world visual stimuli. Unless ... and here is the point of this email ... someone has a nice approach to computer vision that produces polygonal models from 3D scenes! I am moderately sure that such things exist, and I could explore the research area via the usual Googling, but I wonder if someone on this list has more knowledge of this domain than I do, and could simply point me to the good stuff. Thanks, Ben Goertzel ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
