# DAVIDSTUTZ

27thNOVEMBER2015

H. Jégou, M. Douze, C. C. Schmid, P. Pérez. Aggregating local descriptors into a compact image representation. In Computer Vision and Pattern Recognition, Conference on, pages 3304–3311, San Fransisco, California, June 2010.

Similar to the Bag of Visual Words model [1], Jégou et al. approach the problem of image retrieval by first computing a vocabulary $M$ of visual words learned using $k$-means clustering. Instead of counting word occurrences, they consider the corresponding residuals:

$f(y_{l,n}) = (\delta(NN_{\hat{Y}}(y_{l,n}) = \hat{y}_1)(y_{l,n} - \hat{y}_1), \ldots, \delta(NN_{\hat{Y}}(y_{l,n}) = \hat{y}_M)(y_{l,n} - \hat{y}_M))$

where $\hat{Y}= \{\hat{y}_1, \ldots, \hat{y}_M\}$ is the set of learned visual words and $y_{l,n}$ is the $l$-th feature extracted from image $n$. These so-called embeddings are then aggregated using

$F(Y_n) = \sum_{l = 1}^L f(y_{l,n})$.

The image representation $F(Y_n)$ is usually $L_2$-normalized.

• [1] J. Sivic, A. Zisserman. Video google: A text retrieval approach to object matching in videos. In Computer Vision, International Conference on, pages 1470–1477, Nice, France, October 2003.