Content Based Image Retrieval using Bag-Of-Regions: an Efficient Approach

In this work we introduce the Bag-Of-Regions model, in- spired from the Bag-Of-Visual-Words. Instead of clustering local image patches represented by SIFT or related descriptors, low level descriptors are extracted and clustered from image regions, as given by a segmen- tation algorithm. The Bag-Of-Region model allows to de ne visual dic- tionaries that capture extra information with respect to Bag-Of-Visual- Words. Combined description schemes and ad-hoc incremental clustering for visual dictionnaries are proposed. The results on public datasets are promising.