CVPR '96 Tutorials - June 16 and 17, 1996

Tutorial 1: Shape Representations for Object Recognition

Sunday, June 16, 8:00 a.m. - 12:15 p.m.
Benjamin B. Kimia
Brown University

Shape representation plays a key role in object recognition. In this tutorial we will cover several approaches to shape representation. The presentation is organized around a classification of these approaches along three axes:

  1. whether they are based on extrinsic primitives, such as superquadrics, or based on intrinsic geometric measurements, such as the Extended Gaussian Image;
  2. whether they are local, such as those based on measures of curvature, or are global, e.g. those based on moments;
  3. whether they are based on parts or based on deformations.
When appropriate, for each approach we will discuss the issue of how to derive such representations directly from the image, e.g. using active contours. Finally, we will discuss the suitability of each representation under visual transformations such as occlusion, object deformation, movement of parts, changes in the lighting condition, changes in viewing.

Biography

Benjamin B. Kimia is a member of faculty at the Division of Engineering at Brown University. He is also a member of the Laboratory for Engineering Man/Machine Systems (LEMS), an interdisciplinary group focused on signal and image processing, control, and computer engineering. Dr. Kimia received the B.Eng. Honors degree from McGill University, Montreal, Canada in 1982. He subsequently pursued graduate studies there towards M.Eng. (1986) and Ph.D. (1991) in the area of Computer Vision and Image Processing. Prof. Kimia's current research interests are focused on mathematical, psychophysical, and computational models for visual processing, in particular the application of modern PDE and curve/surface evolution methods in computer vision. His research program focuses its investigation on a representation of shape as the singularity set of its deformations and its applicability to recognition.

Tutorial 2: Medical Image Analysis

Sunday, June 16, 1:00 p.m. - 5:15 p.m.
Torfinn Taxt and Arvid Lundervold
Biomedical Imaging Group, University of Bergen, Norway

The tutorial will have a strong emphasis on medical ultrasound imaging and medical magnetic resonance imaging. These two modalities are, except for X-ray imaging, the most important imaging modalities in medicine. In contrast to X-ray imaging modalities, both the ultrasound and MR imaging modalities are going through a dramatic expansion in their application areas and are areas of very intense research and commercial interest.

The tutorial is split in four parts. The first part is an introduction giving relevant background material. The second part is dedicated to medical ultrasound imaging and the third part is concerned with medical magnetic resonance imaging. The fourth and final part concentrates on future trends in medical image analysis with particular emphasis on ultrasound imaging and magnetic resonance imaging.

Biographies

Torfinn Taxt was born in 1950 and is professor in biomedical imaging at the University of Bergen, Norway, and professor in image processing at the University of Oslo. He has the Ph.D. in developmental neuroscience (1983), the M.D. (1976) and the M.S. in computer science (1978) from the University of Oslo. He is associate editor of IEEE Medical Imaging and Pattern Recognition. His primary research interests are restoration, segmentation and multispectral analysis applied to medical ultrasound and magnetic resonance images. He has also published papers on image models, remote sensing and document processing.

Arvid Lundervold has a B.Sc. in mathematics/philosophy (1976), an M.D. (1982) from the University of Oslo, and a Ph.D. in medical image analysis, from University of Bergen (1995). Presently he is associate professor in medical informatics at the University of Bergen, Section for medical image analysis and pattern recognition. His principal research interests are multispectral statistical classification, tissue characterization, 3D imaging, volume visualization, and functional magnetic resonance imaging. Arvid Lundervold has about 30 international publications in neuroscience and medical imaging.

Tutorial 3: Content-Based Retrieval for Image and Video

Monday, June 17, 8:00 a.m. - 12:15 p.m.
Rosalind W. Picard
MIT Media Lab

Content-based retrieval is in increasing demand, with applications spanning consumer imagery, medical research, film-editing, advertising, and more. This tutorial will provide an understanding of the current research problems in this area, focusing on pattern recognition and computer vision solutions for representation and recognition of image and video content. Specifics to be addressed include: techniques for similarity comparisons based on texture, color, shape and combinations of features, methods which operate on compressed data, online learning algorithms for assisting in classification and annotation, online learning of subjective labels and user preferences, and methods for organizing video and image collections to facilitate browsing, retrieval, and annotation. Applications and existing systems will also be discussed.

Biography

Rosalind W. Picard received the B.E.E. from Georgia Tech in 1984 and worked as a Member of the Technical Staff at AT&T Bell Labs from 1984-1987. She earned the M.S. and Sc.D. in Electrical Engineering and Computer Science from the Massachusetts Institute of Technology in 1986 and 1991 respectively, and in 1991 was appointed Assistant Professor at the M.I.T. Media Laboratory. In 1992 she was awarded the NEC Development Chair in Computers and Communications, and in 1995 was promoted to Associate Professor. One of the pioneers in content-based video and image retrieval, Dr. Picard is currently guest editor of the IEEE Transactions on Pattern Analysis and Machine Intelligence special issue on Digital Libraries. Her research interests include pattern recognition and learning, texture and pattern modeling, video and image analysis, and affective computing.

Tutorial 4: Genetic Algorithms and Genetic Programming

Monday, June 17, 1:00 p.m. - 5:15 p.m.
John Koza
Stanford University

This tutorial will introduce participants to the ideas and applications of genetic algorithms and genetic programming.

The genetic algorithm is a domain-independent search technique that solves, or approximately solves, problems based on the mechanics of Darwinian natural selection and genetics. Genetic algorithms are receiving increased attention in many areas, including pattern recognition. The tutorial will introduce the mechanics of a simple genetic algorithm and consider the theory of implicit parallelism that underlies its problem-solving power. A parade of current applications will be reviewed.

Genetic programming is a domain-independent technique for automatic programming that evolves a computer program that solves, or approximately solves, problems.

Genetic programming has found applications in a wide variety of different areas including system identification, control pattern recognition, modeling, design of electrical circuits, forecasting, empirical discovery, data mining, robotics, automatic programming of multi-agent strategies, distributed artificial intelligence, game theory, optimization, and computational aspects of molecular biology.

Biography

John R. Koza is a consulting professor in the Computer Science Department at Stanford University. He is author of two books from the MIT Press on genetic programming: Genetic Programming: On the Programming of Computer by Means of Natural Selection (1992) and Genetic Programming II: Automatic Discovery of Reusable Programs (1994).

Back to CVPR '96 Home Page