Diversification with Fisher Kernel pseudo-feedback

In this article we approach the problem of image search result diversification from a novel perspective that involves the use of relevance feedback (RF). Traditional RF introduces the user in the processing loop by harvesting feedback about the relevance of the search results. This information is used for recomputing a better representation of the data needed. The novelty of our approach is twofold. First, we exploit the RF concept in a completely automated manner via pseudo-relevance feedback; this is while addressing the diversification in priority rather than the relevance. Secondly, we introduce a more efficient visual content representation scheme that exploits Fisher Kernels (FK). It allows to better capture variability of visual keypoints information. We use unsupervised hierarchical clustering to re-group FK descriptors in classes. Diversification is finally achieved with a re-ranking approach. Experimental validation on Flickr data shows the advantages of this approach.