Lapas attēli
PDF
ePub

• Feature extraction & Object This meeting was reasonably contention of innumerable ideas reflects recognition

representative of the state of the art. the direct importance of the problem,

For example, one Korean paper (Ref 1) its attraction to vision researchers as a • Applications

at this meeting reported on a system problem universe of large but controlled

for extracting car types and license plate size, and the lack of conceptual converI would like to express my sincere numbers from camera images, that is, gence in this area. There are so many thanks to

in place and working well in its limited uses for automated literacy that effort

universe of car and plate types. The and resources will continue to pour in Prof. Timothy Poston

problem of workaday character recog- this direction, but it would be unwise at Department of Mathematics nition is a much larger one in East Asia this time to place any bets on what Pohang Institute of Science than in pure-alphabet countries (though method -existing or still to be and Technology

even there decorative scripts, from developed--will finally dominate the P.O. Box 125

Shatter to ornamental Arabic, make a field of character recognition. Pohang, Kyung Buk 790-330, Korea universe too wild and varied for exist- In any meeting about computer Tel: +82-562-79-2052

ing computer methods). A Japanese analyses and decision-making, Fax: +82-562-79-2799

high school graduate is supposed to nowadays, one expects neural networks. E-mail: tim@vision.postech.ac.kr recognize about 2,000 Chinese charac- At this conference there were five: using

ters; a Korean who knows only the networks for identifying objects in an who contributed the following mate- phonetic script is functional but can- image (“Choose one out of Sky/Grass/ rial. Readers should note that many not read (for instance) most newspaper Road/Tree/Road/Car”) (Ref 4), segimportant Japanese research topics on headlines. Identifying characters from menting simple images (“Separate this computer vision were not presented a universe of thousands, even in a fixed stool sketch from the background sketch here.

typeface, is a qualitatively different of floor and folding screen") (Ref 5), It has been said that if all a rat knew problem from working with Western stereo matching (Ref6), an image thinof rat psychology was the information character sets. Just as with English ning method (Ref 7), and a classifier in psychology textbooks, he would fail writing, handwritten text has far more for polyhedra with up to eight faces, at every social interaction he attempted variation and consequent difficulty. Thus most four meeting at a point (Ref 8). with another rat. Similarly, if the pro- to achieve over 98% cumulative accu- As is common for neural net research, cessing of his visual input rested on racy on a realistically large character the problems handled were quite small, current algorithms for vision, he would set is not a small achievement. This was and while directions for development be safest to rely on his excellent sense done by two Japanese papers in radi- were pointed out, there was no analysis of smell. Broadly speaking, most com- cally different ways. One (Ref 2) used of the way the network's necessary size puter vision applications depend on an Fourier transforms of rectangular and learning time would scale with the extremely predictable environment: “is windows within a character to estimate complexity of the problem. In most that a nut or a bolt?” algorithms that how like a diagonal/vertical/etc. stroke network problems, unfortunately, these depend on consistent lighting and would that part of the character seemed, tested scaling properties are abominably bad, , often report“bolt" for a severed finger. on 881 character categories from a so that the network “solution” is no The highly stereotyped behavior of an standard database of handwritten char- better than a classical algorithm that animal adapted to cage life (and no acters. The other (Ref 3) worked on takes exponential time or worse, except longer viable in the wild) is richness the cheap printing in a translated Isaac for the “learning” that reduces the load itself compared to any manufactured Asimov novel (processing it in about on human intelligence in creating the system. Since back-of-the-envelope the time Asimov seems to need to write solution. Some of the papers here may calculations suggest that the process- one), which involved 1,164 distinct scale usefully--some neural networks ing capacity of the current generation characters. This paper used a more are proving useful in practical of supercomputers is up there with the directly geometrical approach, search- applications--but none of them address nervous system of a housefly, it is a ing for pieces of approximate straight the question. remarkable fact that progress is, in line within the image, calculating their The enormous range of methods fact, being made in solving visual tasks lengths, and so on. Many other methods applied to scene analysis (optical flow, far more interesting to humans than are under development (some of which modelling of the object seen and comanything a fly can do.

look unlikely ever to scale to a large parison of the image with prediction, character set with good reliability); this fitting a distorted quadric surface, analysis of a moving 3D outline, shape Since major progress here would be a 3. “An Experiment on Printed from shading...) generously represented large step toward understanding the Japanese Character Recognition using at this meeting represents not only the dynamics of consciousness, it is not a a PC for the Braille Translation of Novel immaturity of the field (as with charac- trivial problem. Not surprisingly, at this Books," Yasuhiro Shimada and Mitsuru ter recognition) but almost certainly meeting there was no session on inte. Shiono, Okayama University of Scithe multifaceted nature of the prob- grating the output of the descriptors ence, Japan. lem. The human vision system can for rigid shapes, faces, etc. discussed in respond "couple dancing!" to a grey- the many papers on how to use camera 4. “Hopfield Net-Based Image Labeltoned image, a line sketch, a half-second images, range data, and so forth. ling with MRF-Based Energy Funcmovie showing only points of light As one might expect, given the respec- tion,” Byoong K. Ko and Hyun S. Yang, attached to dancers in the dark ... and tive research populations and funding KAIST. thus solves its problems in multiple of Japan and South Korea, there were ways. This multiplicity is presumablyin 47 papers from Japan against 33 from 5. "Image Segmentation Using Neural somesense necessary, as the evolution- the host country, of which a certain Networks,” Ao Guo-Li, Cui Yu-Jun, ary cost of evolving it cannot have been number were“trial flights” by graduate Masao Izumi, and Kunio Fukunaga, low. Complicated systems have many students giving their first conference College of Engineering, University of potential defects, so that many muta- papers. In some cases, this was pain- Osaka Prefecture, Japan. tions could cripple them, and very few- fully obvious in the quality of the work at a given moment--improve their pres- as well as in the confidence of the pre- 6. “Stereo Matching Using Neural ent working. The papers here repre- sentation. The experience of placing Network of an Optimized Energy sent normal progress in existing work in the setting of a larger and more Function,” Jun Jae Lee, Seok Je Cho, approaches to subproblems in the Great developed research effort will certainly and Yeong Ho Ha, Kyungbuk National Problem of “What am I seeing?"--a be strengthening for Korean work in University, Korea. number of papers that specialists will computer vision. need to read, but nothing that starts a

7. "Automatic Construction of Image whole new approach, or represents a REFERENCES

Transformation Processes Using Feastep toward the problem created by the

ture Selection Network," Tomoharu multiplicity itself. Given that a robot 1. “Recognition of Car Type and Nagao, Takeshi Agui, and Hiroshi fit to explore a rich environment will Extraction of Car Number Plate by Image Nagahashi, Tokyo Institute of Techalmost certainly need (like the mam- Processing," Dong-Uk Cho, Young- nology, Japan. malian brain) to use many submethods Lae Bae, and Young-Kyu Kang, Sysin visual analysis, how should it inte- tems Engineering Research Institute/ 8. “3-D Polyhedral Object Recognigrate the results? How should the Korea Institute of Science and Tech- tion using Fast Algorithm of Threecomputer/how does the brain repre- nology (SERI/KIST), Korea.

Dimensional Hough Transform and sent the objects about which so much

Partially Connected Recurrent Neural information arrives in conflicting for- 2. “Recognition of Handprinted Network,” Woo Hyung Lee, Sung Suk mats? As each submethod becomes more Characters by FFT,” Tadayoshi Kim, Kyung Sup Park, and Soo Dong powerful, the problem of integration Shioyama and Akira Okumura, Kyoto Lee, Ulsan University, Korea. or “sensor fusion” becomes more urgent. Institute of Technology, Japan.

Appendix

RESEARCH ACTIVITIES AT KOREAN UNIVERSITIES/

RESEARCH INSTITUTES IN COMPUTER VISION

[Data collected by Prof. Joon H. Han, POSTECH, and Prof. Hyun S. Yang, KAIST)

KYUNGBUK NATIONAL UNIVERSITY

A Research Areas

1. Computer Vision (stereo vision, pattern recognition, range image analysis, motion estimation) 2. Image Analysis (restoration, enhancement, edge extraction and thinning, segmentation, data compression) 3. Neural Network (pattern recognition, stereo vision, image analysis)

B. Projects (partial list)

1. 3D object recognition from 2D images
2. Development of shape recognition and synthesis technology by using image processing techniques
3. Integrated circuit (IC) layout pattern recognition by using image processing techniques
4. 3D shape and motion recognition
5. Studies on human vision

C. Facilities (image processing lab)

1.

Color image processing system (IBM PCAT with color image processor, color CCD camera (512 x 512 x 8 bits), color monitor)

2. Pseudo color/BW image processing system (IBM PC/386 with ITI ljl series image processor, IBM PC/AT with

ITEX-PC-Plus, color CCD camera (512 x 512 x 8 bits), B/W monitor)

3. Stereo vision system (IBM PC/AT with FG-100-AT frame grabber, two CCD cameras, B/W monitor)

4.

Laser range scanner system (Technical Arts) (100X scanner, solid state camera (ICIDTEC), RCA monitor, laser power supply, Visual 500 terminal)

5.

SUN 4/260C workstation (color graphic system, color monitor (1280 x 1024), digitizer tablet, plotter)

6. IR Camera
7. Film Recoder
8. Printer

D. Faculty - Prof. Sung Il Chien

SOGANG UNIVERSITY

A. Research Areas

1. Character Recognition
2. Stereo Vision
3. Image Coding

B. Faculty Members - Prof. Kyung Whan Oh, Prof. Rae Hong Park

SEOUL NATIONAL UNIVERSITY

Signal Processing Lab

A Research Areas

1. Image Coding (2nd generation coding, region based coding, texture analysis/synthesis, motion compensated

coding, motion detector, target tracker (real-time)]

2. Computer Vision (low-level segmentation (color, B/W), shape matching (relaxation), polygonal

approximation)

B. Facilities (Gould 8400 IP + 19-inch RGB monitor, Micro-VAX II, Image Technology IP512 Image Processing

System, SNU RGB Image Processing System, PDP 11/23, IBM PC 386, AT, XT)

C. Faculty - Prof. Sang Uk Lee

Automation & Systems Research Institute

A. Research Areas (Computer Vision (low and high level)]

B. Current Projects (On the development of a color vision system employing DSP, Real-time vision system)

C. Facilities (SUN 4 workstations, CCD camera, Adaptive robot, IP 150 Image Processing System, IBM PC/AT, 386

etc.)

D. Researchers - Prof. Sang Uk Lee (SNU), Prof. Jhong Soo Choi (Chung-Ang Univ.), Prof. Rae Hon Park (Sogang

Univ.), 6 research assist.

YONSEI UNIVERSITY

A Research Areas

1. Neural Network Modeling
2. Korean Character Recognition
3. Dynamic Character Recognition
4. Korean Character Processing Computer

B. Facilities (Micro-VAX II, Vax II/750, Solbourne, IBM PC/AT, scanner, B/W camera, printers)
C. Researchers - Prof. Il Byung Lee, -10 graduate students

CHUNG-ANG UNIVERSITY

Image & Information Engineering Lab

A Research Areas

1. Medical Ultrasound Imaging
2. Computer Vision
3. Visual Communication

B. Current Projects (a study on image understanding system using active focusing and meta-knowledge)
C. Facilities (workstations, PCs 132-bit, image data acquisition system, plotter, logic analyzer, IBM 3090)
D. Researchers - Prof. Jong-Soo Choi, 1 assist. prof., 17 graduate students

Computer Vision & Graphics Lab

A Research Areas

1. Computer Vision
2. Image Understanding
3. Pattern Recognition
4. Computer Graphics

B. Projects (partial list)

1. Construction of image understanding system
2. Basic research on image processing
3. Basic research on artificial intelligence

C. Facilities (CCD camera, frame grabber, PCs)
D. Faculty - Prof. Young Bin Kwon

CHOONGBUK UNIVERSITY (AI LAB)

A. Research Areas (Projects)

1. Development of On-Line Handwritten Chinese Character Recognition 2. Evaluation of Image Skeleton Algorithms

B. Facilities (SUN SPARC workstations, Macintosh workstations, VGA color PC 386, VGA color Notebook 386,

WACOM SD-510C tablet digitizer, laser beam printers, IBM PC/ATs)

C. Researchers - Prof. Seong Whan Lee, 5 graduate students, 2 research scientists

« iepriekšējāTurpināt »