Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!ucsd!ucbvax!ADS.COM!Vision-List-Request From: Vision-List-Request@ADS.COM (Vision-List moderator Phil Kahn) Newsgroups: comp.ai.vision Subject: Vision-List digest delayed redistribution Message-ID: <9011150500.AA15063@deimos.ads.com> Date: 14 Nov 90 17:39:46 GMT Sender: daemon@ucbvax.BERKELEY.EDU Reply-To: Vision-List@ADS.COM Distribution: inet Organization: The Internet Lines: 203 Approved: vision-list@ads.com Vision-List Digest Wed Nov 14 09:39:46 PDT 90 - Send submissions to Vision-List@ADS.COM - Send requests for list membership to Vision-List-Request@ADS.COM Today's Topics: Re: Shape from Shading Motion Detection Using Neural Nets Making personal bibliographies public MPEG/JPEG images, software implementations Video image sequences Colour ref. ---------------------------------------------------------------------- Date: Mon, 12 Nov 90 22:26:11 EST From: mancini@jhunix.hcf.jhu.edu (Todd A. Mancini) Subject: Re: Shape from Shading Ikeuchi and Horn make several simplifying assumptions in their original shape from shading algorithm. I am more interested in processing real images, so I have begun my work using ray-traced images. These images have several properties which are not handled well by the original algorithm, such as perspective projection, non-centered objects, and Lambertian models which diminish source intensity as distance increases. My intial models are quadrics (and super-quadrics). Simpler methods of enforcing continuity in the smoothness over the surface given the occlduing boundary work better and faster than Ikeuchi and Horn's method which also seeks out to use intensity information. There are other problems to be dealt with; one involves an over-simplification corresponding surface orientations to expected intensity values given a Lambertian model. In effect, the algorithm presented in the paper treats all objects as spheres (or ellipsoids) of unit size. It is possible to create quadrics which will not render correctly given their simple definition of a reflectance map. I am seeking an algorithm which is more concerned with observed intensity values in the bit-mapped image to find the orientation map. Ideally, such an algorithm would not require the orientation at the occulding boundary to be input, but by means of a system similar to edge detection would quickly determine the occluding boundary and work off of that. I have already started to show that for a large set of smooth surfaces, it is enough to determine the occluding boundary to find the entire surface orientation with a very high degree of accuracy; information from the intensity map over the surface will be mainly used to resolve local perturbations from the smoothness (even in light of image noise.) -Todd ------------------------------ Date: 13 Nov 90 22:56:52 GMT From: leadsv!laic!stiles@decwrl.dec.com (Randy Stiles) Subject: Motion Detection Using Neural Nets Keywords: neural networks, motion detection, neuroethology, machine vision, connectionism Organization: Lockheed AI Center, Menlo Park Fellow netters, Does anyone out there have more info on this development reported last week by the Wall Street Journal, concerning visual object recognition using color to recognize objects and motion detection based on the dragonfly's visual system? The paper reports Michael Swain at U. Chicago and Randal Nelson at U. Rochester, respectively, as the researchers on these systems. I would be especially keen to receive information from them about these developments. Please send any info directly to my email address stiles@laic.lockheed.com. I will gather the messages together and post them to the net. Randy Stiles (stiles@laic.lockheed.com) [ Interesting to see what appears in the popular media. Swain/Wixson/Ballard and Nelson's work is described in the Proceedings of the AAAI-90 Workshop on Qualitative Vision. See also Nelson & Aloimonos, PAMI, Oct. 1989; also University of Rochester TRs. phil... ] Wall Street Journal Friday, November 9, 1990: Scientists are teaching robots and computers to see in living colors and to identify moving objects while in motion themselves. In research funded partly by the Pentagon's Defense Advanced Research Projects Agency (DARPA), scientists at the University of Rochester and the University of Chicago report successfully programming robots to recognize multicolored objects solely by their colors. That's a departure from the traditional approach, in which objects are recognized by their shape, and from machine vision experiments, in which robots "see" only single or dominant colors in a pattern. The Rochester robot, for example, was capable of picking out a box of Kellogg's Frosted Flakes from among 70 similiarly colored objects. And while most robots today must remain still to detect movement, another robot in Rochester was programmed to pick out moving objects while on the go itself. the robot was programmed to identify objects that aren't moving in synch with the movement of its own visual field. Randal Nelson, an assistant progessor of computer science at Rochester, says the software was inspired by the dragonfly, which is though to stalk its prey by identifying small objects whose movement doesn't coincide with that of the rest of the its visual field as it buzzes around. Michael Swain, a University of Chicago researcher, says he has received inquireies from companies that view the color vision research as potentially useful in commercial applications such as supermarket checkout systems. "You can't bar-code a squash," he says, but the computer might one day be programmed to recognize one. Mr. Nelson sees his motion-detection research in future surveillance systems. ------------------------------ Date: Tue, 13 Nov 90 14:10:13 GMT From: cs_s424@ux.kingston.ac.uk Subject: Making personal bibliographies public Organisation: Kingston Polytechnic Hello I would like to thank Patrick Flynn for sending his personal bibliography in to the list. I was able to make use of many references. Perhaps other members of the vision list could do the same? Or send in a description of the key areas in your bibliography so that it can be made available on request. Regards, Paul Netherwood janet : P.J.Netherwood@uk.ac.kingston Research internet : P.J.Netherwood@kingston.ac.uk phone : (+44) 81 549 1366 ext 2923 local : cs_s424@ux.king School of Computer Science and Electronic Systems, Kingston Polytechnic, Penrhyn Road, Kingston-upon-Thames, Surrey KT1 2EE, UK. ------------------------------ Date: Tue, 13 Nov 90 19:35:26 +0100 From: John Husoy-stip Subject: MPEG/JPEG images, software implementations The Joint Photographic Experts Group (JPEG), a working group under ISO/IEC have recently agreed on a standard for the compression of digital still images. The Motion Pictures Expert Group (MPEG) has been concered with the same problem in conjunction with storage of digial video sequences. We are currently engaged in image coding research for both video and still images and would like to test our algorithms on the images/image sequences that have been used in conjunction with the standardization activities. Therefore, -- does anybody know of a source where I can get hold of test images/sequences that has been used by the JPEG/MPEG working groups? Also is anyone aware of any software implementing these coding standards, commerically or otherwise. Any help is greatly appreciated! John Haakon Husoy The Norwegian Institute of Technology Department of Electical and Computer Engineering 7034 Trondheim - NTH NORWAY email: jonh@tele.unit.no tel: ++ 47 + 7 + 594453 fax: ++ 47 + 7 + 944475 ------------------------------ Date: Tue, 13 Nov 90 13:36:47 CST From: silsbee@vision.ee.utexas.edu (Peter Silsbee) Subject: video image sequences Hello, we are doing research in image sequence compression. For evaluation of results, it is necessary to record the compressed/decompressed sequences on videotape. The equipment to which we have access is not great and we are looking for an alternative. We would appreciate any information about what equipment other researchers in the field are using that gives high-quality video. Thanks in advance, Peter Silsbee ------------------------------ Date: Wed, 14 Nov 90 14:30:54 BST From: A.ALLEN@aberdeen.ac.uk Subject: Colour ref. NJC Strachan, P Nesvadba, and AR Allen, "Calibration of a Video Camera Digitising System in the CIE L*u*v* Colour Space", Pattern Recognition Letters, (in press, Nov. 1990). Dr A R Allen Dept of Engineering, University of Aberdeen, Aberdeen, AB9 2UE, UK. ------------------------------ End of VISION-LIST ********************