Séminaire du jeudi 15 avril
Oratrice: Cordélia Schmid
Titre: Aggregating local descriptors into a compact image representation
Résumé: We address the problem of image search on a very large scale, where<tt> </tt>three constraints have to be considered jointly: the accuracy of the<tt> </tt>search, its efficiency, and the memory usage of the representation. We<tt> </tt>first propose a simple yet efficient way of aggregating local image<tt> </tt>descriptors into a vector of limited dimension, which can be viewed as<tt> </tt>a simplification of the Fisher kernel representation. We then show how<tt> </tt>to jointly optimize the dimension reduction and the indexing<tt> </tt>algorithm, so that it best preserves the quality of vector<tt> </tt>comparison. The evaluation shows that our approach significantly<tt> </tt>outperforms the state of the art: the search accuracy is comparable to<tt> </tt>the bag-of-features approach for an image representation that fits in<tt> </tt>20 bytes. Searching a 10 million image dataset takes about 50ms.
This is a joint work with H. Jegou, M. Douze and P. Perez
<tt>
</tt>
<tt> </tt>
 |
CMLA UMR 8536 |
 |