2008/4/24 Mike Tintner <[EMAIL PROTECTED]>: > Thanks for reply, but you haven't quite answered. Geometry, in my > definition, is a systematic set of regular forms, which can be, and are, > used to deconstruct visual forms in various kinds of images.
This implies that the image contains sufficient information which can then be converted into knowledge about objects, etc. Unfortunately in real situations this often isn't the case. Understanding images is at least as much about reconstruction as it is about deconstruction. And geometry is > , as I understand - please correct me - the whole basis of current > approaches to visual object recognition. I don't think that will work, > because the "meaning" or, perhaps better, "sense", of the object - of the > dancer in that picture, say, - lies in the whole form, the whole shape of > the object or dancer. And as soon as you deconstruct it, you've lost it. The > shape can only be processed in an understandable way as a whole shape. (And, > if that's correct, the perceptual brain is a cartoonist, rather than a > geometer). You can think about an object as a constellation of features having certain geometric relationships to each other. This is basically how the 2D forms of object recognition in systems like the ones produced by Evolution Robotics and Skilligent work, and these perform reasonably well on objects with planar surfaces. What I think needs to be done is to extend this into 3D and eventually 4D, so that each feature is a locally oriented plane having a surface normal (see Andrew Davison's MonoSLAM system for an example of this). This would mean that objects with complex shapes could be recognized from arbitrary viewing perspectives after only minimal training. Competence in this type of vision may also be closely related to the simultaneous localization and mapping problem. For an example of existing 2D object recognition see http://www.skilligent.com/products/computer-vision.shtml This is really all about computational geometry. ------------------------------------------- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244&id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
