Visual object retrieval by graph features

Our approach addresses the well-known problem of adding spatial information to the standard bag-of-visual-words (BoVW) approach. To that end, a set of graphs descriptors are constructed for the entire image or for local image regions. These descriptors are then incorporated into a bag-of-visual-words approach, leading to a method called bag-of-bag-of-visual-words (BBoVW).