Vision-List-Request@ADS.COM (Vision-List moderator Phil Kahn) (04/17/91)
VISION-LIST Digest Tue Apr 16 13:23:14 PDT 91 Volume 10 : Issue 17 - Send submissions to Vision-List@ADS.COM - Send requests for list membership to Vision-List-Request@ADS.COM - Access Vision List Archives via anonymous ftp to ADS.COM Today's Topics: Re: stereo ground truth RE: Graphics ==> Modelling <== Vision Texure Image w/ Ground Truth Sought Handwritten character database Image Compression algorithms IRI D256 Image Processing System - Need Help FEX and LPEG Machine/computer vision for measurement LABOIMAGE v 3.1 available by ftp ---------------------------------------------------------------------- Date: Thu 4 Apr 91 15:50:37-PST From: Marsha Jo Hannah <HANNAH@ai.sri.com> Subject: Re: stereo ground truth Michael McCool <mccool@dgp.toronto.edu> writes: (regarding verified disparity maps, i.e. ground truth, for stereo pairs) > Why can't artificial images be used? Good reflectance models are now > available in computer graphics. "Photorealistic" images, i.e. images > with a large amount of complexity, can now be generated. It might be > argued that such images would not be as good as a test as real images, > yet random-dot stereograms are commonly used as test cases, and are > blatantly artificial. Use of computer-generated imagery would allow > precise control over image features and camera geometry, all without > building photography jigs. AND the ground truth would be trivially > available. Artificially generated stereo imagery does provide one way to test stereo algorithms. The problem is that no model can be as complex as is the real world as seen by a real camera. Artificial images tend to be a little too smooth on surfaces and a little too jagged on depth discontinuities. Image generation keeps getting better, but it still isn't "real". There is also the problem of believability. It is possible for artificial imagery to be designed in a manner that a given algorithm will do either much better or much worse than it would on real imagery. Sometimes this is done on purpose, to show off how good one's algorithm is and how bad the competition is. More often, it is unintentional, resulting from simplifying assumptions on the part of a researcher as to what the world really looks like---his image generator is likely to use the same type of world model as his stereo algorithm. Most of the questions that I have heard about ground truth were posed by people who fund research, and most of them were decidedly leary of proposals to do the evaluation on artificial imagery. The ISPRS image matching data set did contain one image pair that was generated---a single image (which resembled a blurry random-dot stereogram) was projected onto a mathematical surface, thence to a second artificial camera. Everyone could do this image pair fairly well, so it didn't really tell the test architects much. The other 11 real image pairs went a lot further toward pointing up which algorithms could handle what. mjh ------------------------------ Date: Fri, 5 Apr 1991 08:58:18 +0200 From: Thierry Pun <pun@cui.unige.ch> Subject: RE: Graphics ==> Modelling <== Vision There are more and more useful combinations of computer vision and graphics, in various subfields of CG as well as CV. In particular, I am aware of several projects that have been going for some time in the direction of what you called "cosmic derivative". The Eurographics association has a working group on this topic of interactions between image synthesis and analysis. A paper has been published which overviews these various relationships between the two domains. For example, in terms of common theoretical basis and concepts, the following topics are discussed: perception, light, quasi-linear paradigm, stochastic paradigm, geometry, physics, mathematics, computer science. Also addressed are common areas of study and common applications (T. Pun and E. Blake, "Relationships between image synthesis and analysis: towards unification?", Computer Graphics Forum, 9, 2, July 1990, 149-163). Thierry Pun ------------------------------ Date: Thu, 11 Apr 91 14:25:05 CDT From: jdm5548@diamond.tamu.edu (James Darrell McCauley) Subject: Texure Image w/ Ground Truth Sought I've coded up the algorithms to find textural features from: Haralick, R.M., K. Shanmugam, and I. Dinstein. 1973. Textural features for image classification. IEEE Transactions on Systems, Man, and Cybertinetics, SMC-3(6):610-621. I am looking for an image to test them on. For those of you not familiar with that paper, he uses a cooccurence matrix to find: 1 "Angular Second Moment " 2 "Contrast " 3 "Correlation " 4 "Variance " 5 "Inverse Diff Moment " 6 "Sum Average " 7 "Sum Variance " 8 "Sum Entropy " 9 "Entropy " 10 "Difference Variance " 11 "Difference Entropy " 12 "Meas of Correlation " 13 "Meas of Correlation " 14 "Max Correlation Coeff " I only need to see if my coding is correct. Any pointers appreciated. If you have *anything* close to this, please send it to me. I've just about exhausted all other avenues (short of buying an image) and am desperate (and short on time). Thanks, James Darrell McCauley, Grad Res Asst, Spatial Analysis Lab Dept of Ag Engr, Texas A&M Univ, College Station, TX 77843-2117, USA (jdm5548@diamond.tamu.edu, jdm5548@tamagen.bitnet) ------------------------------ Date: Wed, 10 Apr 91 19:44:26 GMT From: drolet@wiener.ino.qc.ca (Jean-Jacques Drolet) Subject: Handwritten character database Organization: National Optics Institute, Quebec, Canada The National Optics Institute of Canada plans to study pattern recognition problems involving handwritten characters. Does anyone know where we could find a large database of digitized handwritten characters? Many thanks! Jean-Jacques Drolet | Snail: 2631 boul. Liegeois, Sainte-Foy National Optics Institute | Quebec, Canada, G1W 1Z5 Phone: +1 418 657 7006 | Internet: drolet@drolet.CAM.ORG Home phone: +1 418 651 3796 | UUCP: uunet!drolet!drolet ------------------------------ Date: Mon, 25 Mar 91 13:47:53 EST From: perry@dewey.CSS.GOV (John Perry) Subject: Image Compression algorithms Take a look at the JPEG ISO imaging compression standard that will be completed by this fall. Joint Photographics Experts Group ISO/IEC JTC1/SC2/WG8 CCITT SGVIII It uses frequency domain transformations (Adaptive Discrete Cosing Transform) to reach lossless compression levels of at least 20:1 and decimated compression of 40:1. If your looking for something simpler, there are lots of public domain tools to reach a compression of 2:1 or so. Several companies are working on chips to speed the JPEG up. We have almost got our software implementation finished for the DECstation and Sparkstation. Mail if you need more.... Richard Hubert Interactive Objects Software GmbH (iO) Nikolausstrasse 20 7807 Elzach, Germany Tel: (+49)-7682-6375 or -6374 FAX: (+49)-7682-6375 (yes, it's an automatic FAX/Telephone) Email: hubert@iobj.uucp Email from DEC VAXmail: decum::"hubert%iobj.uucp@unido.uucp" Email long forms: Internet: hubert%iobj.uucp@unido.informatik.uni-dortmund.de uucp: uunet!unido!iobj!hubert uucp: ...!{decvax,ncar,purdue,rutgers}!unido!iobj!hubert ------------------------------ Date: 9 Apr 91 22:10:20 GMT From: dsmith@eecs.cs.pdx.edu (Guess Who) Organization: Portland State University, Portland, OR Keywords: image processing vision Subject: IRI D256 Image Processing System - Need Help Hello, We have an IRI D256 image processing system. IRI sold a few of these systems before they went out of business. During a move we lost a piece of documentation for the color retrofit board. The color retrofit board comes with a non-sync RGB camera. I know that there is a set up command at the debug level and some iUNIX commands which are required for the system and the monitor to work correctly without a camera sync signal. We lost the list of commands. If anyone knows the command sequence, we would appreciate the information. David Smith, EE Dept., Portland State University ------------------------------ Date: Fri, 05 Apr 91 22:05:49 +0100 From: A.Etemadi@ee.surrey.ac.uk Subject: FEX and LPEG I have written a couple of packages which maybe of interest to people. FEX:: is a package for extracting straight lines and curves from an edge detected image. The edges should first be converted to ASCII strings of pixels. A utility written by Rosin and West is available for performing this opertation. FEX has only been tested on data from 256x256 images. It is very simple to use. There are no thresholds to set. I Generally find that I can reconstruct the Canny edge map analytically. The execution time is on average better than 0.05 seconds per string on a Sun 4 Sparc. FEX may also serve for compressing edge data for transmission since you generally get better than an order of magnitude data compression factor. Finally it also allows the detection of points of high curvature within the edgemap. LPEG:: is a package for perceptual grouping of straight line segments It accepts FEX output directly. There is also a simple routine for the converting the end points of a line segment to the format accepted by LPEG. The output is in the form of ASCII lists of the following: Overlapping Parallel lines Non-overlapping parallel lines Collinear lines L and V junctions T and Lambda junctions The execution time is between 0.6 and 120 seconds on a Sun4 Sparc for lists of between 40 and 150 line segments (actually the time is mostly taken up by IO). Both packages are available by request for evaluation purposes only and come with the usual speel about waranties and distribution. These packages are NOT for distribution to companies however interested parties should contact me directly. Looking forward to seeing you all at the BMVC in Glasgow. ciao Ata <(|)>. Dr. A. Etemadi, | Phone: (0483) 571-281 Ext. 2311 V.S.S.P. Group, | Fax : (0483) 300-803 Dept. of Electronic and Electrical Eng.,| Email: University of Surrey, | Janet: a.etemadi@ee.surrey.ac.uk Guildford, | ata@c.mssl.ucl.ac.uk Surrey GU2 5XH | SPAN : ata@mssl United Kingdom | ata@msslc ------------------------------ Date: 06 Apr 91 16:09:18 EDT From: ZHAO@umde.dbrn.umich.edu Subject: Machine/computer vision for measurement Dear Net:comp.ai.vision Readers: I am asked to compile a report on machine/computer vision techniques/ algorithms for industrial measurements. I appreciate your response to this inquiry to the following specifics: 1. The methodologies used for industrial tool measurement by machine/ computer vision and imaging technology. 2. The current available instrumentations: products using machine/computer vision and imaging technology for industrial measurement. 3. Current research/development efforts. Please send your response to the e-mail address: zhao@umde.dbrn.umich.edu (or to : dzhao@caip.rutgers.edu) Thank you. Sincerely, Dongming Zhao ECE Dept. Univ. of Michigan-Dearborn Dearborn, MI 48128-1491 zhao@umde.dbrn.umich.edu ------------------------------ Date: Tue, 9 Apr 1991 09:49:30 +0200 From: Alain Jacot-Descombes <jacot@cui.unige.ch> Subject: LABOIMAGE v 3.1 available by ftp [ LaboImage is now avilable in the Vision List SHAREWARE archives. (Additionally, the newest version of NIH Image for the Mac has been added.) Hope this code is of interest and use to you... phil... ] LABOIMAGE Original notice T. Pun, March 8th, 1989 (LaboImage 2.0) Updated April 5th, 1989 Updated September 1st, 1989 Updated March 8th, 1990 Updated August 24th, 1990 (new version LaboImage 3.0) Updated March 19th, 1991 (new version LaboImage 3.1) Computer Science Center, University of Geneva, Switzerland Thank you for your interest in Labo Image! GENERAL DESCRIPTION Labo Image is a window based software for image processing and analysis. It contains a comprehensive set of operators as well as general utilities. It is designed to be open-ended; new modules can easily be added. The software is written in C, and currently runs on Sun 3/xxx, Sun 4/xxx (OS3.5, 4.0 and 4.0.3) under SunView. The expert system for image segmentation is written in Allegro Common Lisp. LaboImage has been extensively used by students as well as researchers from various domains: computer science (image analysis), medicine, biology, physics. It is distributed free of charge (source code). STATUS Version 0 has been released in January 1988, version 1 in November 1988, version 2 in May 1989, version 3.0 in August 1990, version 3.1 in March 1991. - hosts: Sun 3/xxx, Sun 4/xxx; - OS: Sun OS 3.5, 4.0, 4.0.3; - window system: SunView, X11/MOTIF as soon as possible; - language: C (Allegro Common Lisp for the expert system); - approx. code size: source 2MB (70'000 lines), executable 2MB under SunView/OS4.0.3; - documentation: manuals (english) MEANS OF DISTRIBUTIONS LaboImage source code is available by anonymous ftp at ads.com, login name anonymous, in pub/VISION-LIST-ARCHIVE/SHAREWARE. If you have no access to ftp, please contact the author. If you wish to be kept current with update, error reports, ..., please send us a mail with your full name, regular and electronic addresses. DISTRIBUTION POLICY In essence: - this is a non-profit software, but it is our property and the copyright notice must appear; - the reference to cite in case of published results obtained with Labo Image is: "A. Jacot-Descombes, K. Todorov, D.F. Hochstrasser, C. Pellegrini and T. Pun: `LaboImage: a Workstation Environment for Research in Image Processing and Analysis', Computer Applications in the Biosciences, 7(2), IRL Press Limited, 1991"; - no responsibility is assumed; - not to be used for profit making purposes; - bugs will usually be corrected since we use intensively the software; - modifications should be communicated to us, with (normally) allowance for redistribution. PAYMENT Athough LaboImage has undergone many upgrades and suffered in the hands of many users, the current version is certainly not bug free. For the time being, we require NO prepayment, return postage or anything. We may however change this policy in the future, and ask for nominal fees to cover material expenditures. HOWEVER, if you are satisfied with the product, why not send us some "souvenir" (edible or not) from your country...??!! CAPABILITIES Labo Image is an interactive software, whose interface is menu, mouse and windows based. It can work on monochrome (binary) or color workstations. Its main features are: - input-output: LaboImage format file, SUN raster file, screen, postscript; - display: mono, RGB, dithering, 3-D perspective display, color table editor; - preprocessing: filters (median, high pass, low pass: hamming, gauss, etc), background subtraction, histogram equalization; - processing: thresholding, Fourier transforms, edge extractions: various operators, ridge-riding, zero-crossing; segmentation into regions, binary and gray tone mathematical morphology; - measures: histograms, image statistics, power, region outlining, object counting; - auxiliary: conversions, arithmetic & logical operations, noise addition, image generation, magnification, convolution/correlation with masks, image; padding; - elementary interactive operations: region outlining, statistics and histogram computation, etc.; - special tools: - modify image at pixel level interactively, - one-dimensional gel analysis, - expert system for image segmentation, - macros definitions, save and replay; - on line documentation. IMAGE FORMATS Own format: descriptor file + data file (binary, byte, int, float, complex; mono or RGB). Supports also Sun raster format. Conversions to various other formats. Constructs: - iconic (pixel-based), with each image having its own parameter list; - vectors (histograms, look-up tables); - graphs (for regions; being implemented); - strings (for macros). MISCELLANEOUS REMARKS (answers to commonly asked questions) - FILE FORMAT: we decided to go for: 1) a machine independant format; 2) a simple, data (ie. signal) oriented format. At the beginning of the development (summer 1987), we were not aware of any image format used by the whole community. There seems now to be some progress on the matter (TIFF, etc.), but they are still not that widely used in the community. Also, due to development priorities we consider conversion routines a more secondary issue as long as our format is simple. In addition, the menu ACQUISITION/LTS is fairly versatile. Also, the SUN raster images can now be read into LaboImage and likewise images on system may be stored in SUN raster format. However.. we would welcome any software contribution! - SUN 3 and SUN 4: Labo Image currently runs on SUN 3/xxx and SUN 4/xxx under SunOS 4.0.3. As long as you can compile on these machines, it should run. - 3D IMAGE PROCESSING: nothing special for such images. - REGIONAL DESCRIPTION: work is underway to develop region segmentation algorithms. These regions will be described by a graph data structure. - ON LINE HELP: available! ACKNOWLEDGEMENTS More than 10 people have so far participated in this project, and their contribution is gratefully acknowledged. Staff: Pierre-Yves Burgi, Claudia Coiteux-Rosu, Ziping Hu, Alain Jacot- Descombes, Rene Lutz, Christian Pellegrini, Thierry Pun, Marianne Rupp, Krassimir Todorov. Students: Anne Bobillier, Alain Brunner, Markus Buchi, Christian Girard, Rene Perrier, Vrinda Shukla. Amongst them, Ziping Hu is responsible for the expert system for image segmentation, A. Jacot-Descombes is responsible for general design issues, and is the keystone for implementation; R. Lutz is responsible for display manipulations (Color Table Editor,etc.); T. Pun is responsible for the original layout and general design issues; V. Shukla is responsible for the upgrade from LaboImage 2.0 to LaboImage 3.0. CONTACTS Particular problems will be redirected to relevant persons, but we prefer that all communications be made to the same address: e-mail: "pun@uni2a.unige.ch" or pun@cgeuge51.bitnet (if this fails, "pun@cui.unige.ch"). tel.: +(4122) 787 65 82 (T. Pun), 787 65 84 (A. Jacot-Descombes). fax: +(4122) 735 39 05. postal address: Thierry Pun LaboImage Computing Science Center, University of Geneva 12, rue du Lac CH - 1207 Geneva SWITZERLAND ------------------------------ End of VISION-LIST digest 10.17 ************************