The NBNN Kernel

TitleThe NBNN Kernel
Publication TypeConference Paper
Year of Publication2011
AuthorsTuytelaars, T., Fritz M., Saenko K., & Darrell T.
Page(s)1824-1831
Other Numbers3231
Abstract

Naive Bayes Nearest Neighbor (NBNN) has recentlybeen proposed as a powerful, non-parametric approach forobject classification, that manages to achieve remarkablygood results thanks to the avoidance of a vector quantizationstep and the use of image-to-class comparisons, yieldinggood generalization. In this paper, we introduce a kernelizedversion of NBNN. This way, we can learn the classifierin a discriminative setting. Moreover, it then becomesstraightforward to combine it with other kernels. In particular,we show that our NBNN kernel is complementary tostandard bag-of-features based kernels, focussing on localgeneralization as opposed to global image composition. Bycombining them, we achieve state-of-the-art results on Caltech101and 15 Scenes datasets. As a side contribution, wealso investigate how to speed up the NBNN computations.

URLhttp://www.icsi.berkeley.edu/pubs/vision/NBNNkernel11.pdf
Bibliographic Notes

Proceedings of the International Conference on Computer Vision (ICCV), Barcelona, Spain, pp. 1824-1831

Abbreviated Authors

T. Tuytelaars, M. Fritz, K. Saenko, and T. Darrell

ICSI Research Group

Vision

ICSI Publication Type

Article in conference proceedings