Events & News
CS/ECE Joint Talk
| Category | Lecture |
| Date | Friday, February 6, 2009 |
| Time | 10:00 am |
| Location | SU Rm 411-A |
| Details | This is our first meet this semester of the ECE/CS seminar. Room: Career Services Room 411-A. (West side of the 4th floor of the student union building) |
| Speaker | Kobus Barnard |
| Title | Associate Professor |
| Affiliation | Computer Science Department, UA |
Learning Models for Object Structure
The form of objects, organisms, and organs is often closely linked to their
function and identity. Being able to capture and quantify their
three-dimensional structure has many important applications. Examples include recognizing novel instances of objects in a class, reasoning about object utility, automated species identification, and classifying the phenotype of cells with different genotypes grown under different conditions.
In this talk I will argue that the current trend in machine vision of treating object recognition as a two dimensional pattern matching task ignores much of what recognition should mean. In particular, representing, characterizing, and exploiting the three dimensional geometry of the object under study is not addressed. I will then outline a general approach to modeling structure, and simultaneously inferring structure parameters and imaging system parameters using Bayesian methodology. The inference approach applies equally well to fitting an existing model to an individual, as well as to learning the model itself through estimating meta-parameters for a class from multiple examples. Fitting the learned models to new data instances can then be used both for identification and to characterize geometry. I will illustrate the approach in several applications including learning object category structure models for simple furniture objects (e.g. tables and chairs), and fitting a model based on a stochastic L-system to to microscopic fungus from the genus Alternaria.
(Joint work with Barry Pryor (plant pathology), Joseph Schlecht, Ekaterina
Spriggs, and Kyle Simek).
Biography
Kobus Barnard is an associate professor of computing science at the University of Arizona. He received his Ph.D. in computer science from Simon Fraser University in the area of computational color constancy. He then spent two years at the University of California at Berkeley as a post doctoral researcher working on image understanding and digital libraries. His current interests include stochastic modeling approaches to machine vision problems and applications to scientific data.