On the burstiness of visual elements

Burstiness, a phenomenon initially observed in text re- trieval, is the property that a given visual element appears more times in an image than a statistically independent model would predict. In the context of image search, bursti- ness corrupts the visual similarity measure, i.e., the scores used to rank the images. In this paper, we propose a strat- egy to handle visual bursts for bag-of-features based im- age search systems. Experimental results on three reference datasets show that our method significantly and consistently outperforms the state of the art.

• Type: Conference
• Year: 2009
• Where: CVPR
• Authors: Herve Jegou (INRIA), Matthijs Douze (INRIA), Cordelia Schmid (INRIA)
• Keywords: Burstiness, Computer Vision, CVPR 2009, Feature Extraction, Visual Elements