Bundle min-Hashing

We present a feature bundling technique based on min-Hashing. Individual local features are aggregated with features from their spatial neighborhood into bundles. These bundles carry more visual information than single visual words. The recognition of logos in novel images is then performed by querying a database of reference images. We further present a WGC-constrained ransac and a technique that boosts recall for object retrieval by synthesizing images from the original query image or reference images. We demonstrate the benefits of these techniques for both small object retrieval and logo recognition. Our logo recognition system clearly outperforms the current state-of-the-art with a recall of 83 % at a precision of 99 %.