Matt Boutell's Research and Selected Publications

Summary: My primary interest is in understanding of digital home photographs, especially semantic scene classification. Most of my research involves using various types of context. I have also worked on determining image orientation, and on general issues involved in learning by example. More broadly, I am interested in image understanding, computer vision and computer graphics. Over the past few years, my primary collaborators have been my co-advisors, Chris Brown of the University of Rochester and Jiebo Luo of Eastman Kodak Company.

Scene Classification

While unconstrained object recognition is an open problem, how can we extract enough information about a scene to classify it robustly as a beach scene, a sunset, or a party scene? For my master's thesis, I surveyed the state of the art (in 2002). Portions appear in the following URCS Technical Report. We also wrote a paper comparing semantic features (such as the output from sky, grass, and other detectors) to low-level features.

Using Context for Scene Classification

My primary contribution to semantic scene classification is demonstrating the utility of context for classifying scenes. Each of these papers investigates using some type of context, whether temporal (from image collections), spatial (relationships between regions in the same scene), or camera (metadata tags about scene capture properties, such as exposure time and subject distance).

Learning by Example Effectively

In these papers, we extend learning by example in two areas: handling multilabel scenes in training and testing, and using image transforms to recompose images, both for training and testing.

Image Orientation

Automatically determining the orientation of home photos is a difficult problem. The first paper documents a study of the accuracy of human observers on the problem and the cues they use. The second describes a vision system that integrates both low-level and semantic cues to solve the problem.

Copyright notice

This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.

Return to main page