November 3rd, 2011 @ 1130am Ira Kemelmacher (University of Washington)


People are photographed thousands of times during their lifetimes. Taken together those photos form a nearly complete and invaluable visual record, which we call a “photobio”. It samples the appearance space of an individual, capturing variations due to age, facial expressions, pose, reflectance, hairstyle, and so forth. With the advent of photo sharing tools like Facebook coupled with face tagging and recognition technology it becomes very easy to amass huge numbers of photos and we will have access to increasingly complete photobios.

This record, however, is massively disorganized, unlabeled and untapped. In this talk, I will explore ways to match, visualize and model faces from unorganized and uncalibrated image sets such as those found on the internet and in personal photos collections. In particular, I will describe our “Exploring Photobios” work, an approach for generating face animations by computing an optimized, aligned subsequence. The “Face Movies” feature of Google's Picasa provides an implementation of our method targeted to personal photos. I'll also describe work on reconstructing 3D models of faces from Internet photo collections.

seminars/seminaritems/2011-11-03.txt · Last modified: 2011/10/28 14:31 by silberman