Abstracts of Presentations

Three Decades of Texture Analysis

(Html Format) or (MS PowerPoint Format)

Rama Chellappa, Ph.D.

Professor

Department of Electrical and Computer Engg.

Institute for Advanced Computer Studies

University of Maryland, College Park, MD

Over the last thirty years or so, texture synthesis, classification and segmentation has been studied using human and machine vision perspectives. The three decades have witnessed a slow but steady evolution of techniques ranging from co-occurrence matrices to Markov random fields to wavelet packets. In this talk, we will present a critical analysis of key approaches from the 70's, 80s and the nineties. Importantly, we will point out the accomplishments and disappointments.

Robust object detection using scale space filtering

(Html Format) or (MS PowerPoint Format)

Arun Sood, Ph.D.

Professor

Department of Computer Science

George Mason University

Medical image segmentation

(Hyperlink under construction)

William M. Wells III, Ph.D.

Assistant Professor

Department of Radiology

Harvard Medical School and

Brigham and Women’s Hospital

This talk will provide an overview of applications of segmentation in medical image processing. Background material and recent developments will be presented, with an emphasis on the segmentation of Magnetic Resonance images. Examples will be drawn from image-guided surgery, multiple sclerosis image analysis, and other areas.

Segmentation and analysis of range images

(Html Format) or (MS PowerPoint Format)

A. Lynn Abbott (Ph.D.) and Hussein S. Taha

Associate Professor

The Bradley Department of Electrical and Computer Engineering

Virginia Tech

Blacksburg, Virginia

A range image is a dense grid of distance values. This presentation will describe recent results that concern the automatic extraction of three-dimensional (3D) solid objects from a single range image. The work focuses on polyhedral objects, for which some of the visible surfaces may be partially occluded. The detection of roof and step edges leads to an initial segmentation of the image, and this is refined using an iterative split-and-merge approach. An analysis of two-dimensional (2D) region shape and corner proximity then leads to estimates of 3D vertex locations for the objects. These initial estimates are often quite inaccurate, however, particularly near depth discontinuities. Vertex location

estimates can be improved substantially through rule-based analysis coupled with a more traditional optimization process. The resulting system therefore utilizes geometric constraints to guide an optimization process by which object descriptors (in this case, vertices and object faces) are extracted from a dense grid of range data. A byproduct of the analysis is an improved segmentation of the image.

Morphological Techniques for Object detection and Image segmentation

(Html Format) or (MS PowerPoint Format)

John Goutsias, Ph.D.

Professor

Center for Imaging Science and

Department of Electrical and Computer Engineering

The Johns Hopkins University

Baltimore, MD 21218

Mathematical morphology is considered to be a powerful nonlinear tool for image processing and analysis and has been used in numerous applications, including industrial inspection, automatic target detection, biomedical imaging, and remote sensing, just to mention a few.

In this talk we discuss the problems of binary and grayscale morphological image reconstruction and image segmentation. In mathematical morphology, image reconstruction is the process of extracting desirable parts from a given image, which have been marked by a set of markers. Image reconstruction turns out to be very effective in problems of object detection and image segmentation. The problem of binary and grayscale image segmentation by means of morphological operators and, in particular, by means of the watershed transform, is also discussed. Examples illustrate the effectiveness of the discussed techniques in a number of applications, including mine detection, object detection and tracking in FLIR video images, and medical imaging.

Last Updated 11/5/1999 (Friday)