Names and Faces
This work finds correspondences between names in captions and faces in the associated news photographs. The result is a large collection of automatically labeled faces that is extremely varied compared to current face recognition datasets. This work takes advantage of two successes in artificial intelligence research: frontal face detection in images, and proper name detection in text. The problem is then to recognize which face goes with each name. This is easier than general face recognition because there are only a small number of potential identities for each face. As a byproduct of solving the assignment problem between faces and names a model for words surrounding names in captions is produced that is surprisingly effective at predicting what names in a caption will be pictured in the associated photograph.
Automatically Produced Face DictionaryFor more information go to Tamara Berg's page.
Papers
Tamara L. Berg, Alexander C. Berg, Jaety Edwards, Michael Maire, Ryan White, Yee Whye Teh, Erik Learned-Miller, David A. Forsyth
In submission
Tamara L. Berg, Alexander C. Berg, Jaety Edwards, David A. Forsyth
NIPS 2004
Tamara L. Berg, Alexander C. Berg, Jaety Edwards, Michael Maire, Ryan White, Yee Whye Teh, Erik Learned-Miller, David A. Forsyth
Computer Vision and Pattern Recognition (CVPR) 2004, Washington D.C., pp 848-854.