Training image classifiers with similarity metrics, linear programming, and minimal supervision
Abstract
Image classification is a classical computer vision problem with applications to semantic image annotation, querying, and indexing. Recent and effective generative techniques assume Gaussianity, rely on distance metrics, and estimate distributions, but are unfortunately not convex nor keep computational architecture in mind. We propose image content classification through convex linear programming using similarity metrics rather than commonly-used Mahalanobis distances. The algorithm is solved through a hybrid iterative method that takes advantage of optimization space properties. Our optimization problem uses dot products in the feature space exclusively, and therefore can be extended to non-linear kernel functions in the transductive setting.
Additional Information
© 2012 IEEE. We would like to thank Andrew Bolstad at MIT Lincoln Laboratory for all the help, advice, and good ideas he has given us in regard to convex optimization for sparse regularization techniques. This work is sponsored by the Assistant Secretary of Defense for Research & Engineering under Air Force Contract # FA8721-05-C-0002. Opinions, interpretations, conclusions and recommendations are those of the author and are not necessarily endorsed by the United States Government. We would like to thank Andrew Bolstad at MIT Lincoln Laboratory for all the help, advice, and good ideas he has given us in regard to convex optimization for sparse regularization techniques.Additional details
- Eprint ID
- 94485
- DOI
- 10.1109/acssc.2012.6489386
- Resolver ID
- CaltechAUTHORS:20190404-161219565
- Air Force Office of Scientific Research (AFOSR)
- FA8721-05-C-0002
- Created
-
2019-04-05Created from EPrint's datestamp field
- Updated
-
2021-11-16Created from EPrint's last_modified field