A Family of Contextual Measures of Similarity between Distributions with Application to Image Retrieval

During this talk, I will introduce a novel family of contextual measures of similarity between distributions: the similarity between two distributions q and p is measured in the context of a third distribution u. In our framework any traditional measure of similarity / dissimilarity has its contextual counterpart. Especially, for two important families of divergences (Bregman and Csiszar), the contextual similarity computation consists in solving a convex optimization problem. We will focus on the case of multinomials and explain how to compute in practice the similarity for several well-known measures. These contextual measures are then applied to the image retrieval problem. In such a case, the context u is estimated from the neighbors of a query q. One of the main benefits of this approach lies in the fact that using different contexts, and especially contexts at multiple scales (i.e. broad and narrow contexts), provides different views on the same problem. Combining the different v...

• Type: Conference
• Year: 2009
• Where: CVPR
• Authors: Florent Perronnin (Xerox Research Centre Europe), Yan Liu (Xerox Research Centre Europe), Jean-Michel Renders (Xerox Research Centre Europe)
• Keywords: Computer Vision, CVPR 2009, Image Retrieval, Similarity Between Distributions