Face Recognition in Totally Unconstrained Environments
Speaker: Erik Learned-Miller , Assistant Professor, University of Massachusetts, Amherst
Date: April 16 2008
Time: 3:00PM to 4:00PM
Location: Star Seminar Room (32-D463)
Host: C. Mario Christoudias, Gerald Dalley, MIT CSAIL
Contact: C. Mario Christoudias, Gerald Dalley, 3-4278, 3-6095, cmch@csail.mit.edu, dalleyg@mit.edu
Relevant URL: Most face recognition papers still address the problem of carefully controlled face recognition. In typical face recognition databases, most of the parameters such as background, lighting, pose, expression, and camera are controlled, and either not varied at all, or varied in a systematic fashion over a small number of values.
Our recent work looks at the other end of the spectrum, face recognition and processing in completely unconstrained environments. We believe addressing this significantly harder problem will push recognition and modeling in new directions, and ultimately is necessary to achieve practical performance for many real-world applications. In particular, we believe studying this problem will force us to use substantially more sophisticated methods, fusing a variety of techniques from low-level vision, mid-level vision, and a variety of machine learning methods. We report on the start of the art in unconstrained face recognition (not very good!) and report a variety of results on recognition itself, and related subproblems such as alignment, pose estimation, and face segmentation.
See other events that are part of MIT Machine Vision Colloquium 2007/2008
See other events happening in April 2008
