Universiteit Leiden Opleiding Informatica Geometry-preserving algorithms for Content-based Image Retrieval Geometry-preserving algorithms for content-based image retrieval

Abstract Content-based image retrieval algorithms are frequently based on the bag-of-visualwords (BoV) representation of images. The BoV representation discards all spatial information. In this paper, we propose an algorithm which encodes spatial information into the BoV representation. Our algorithm forms geometry-preserving visual phrases which are invariant to translation and rotation. Furthermore, we demonstrate that our algorithm outperforms other recently proposed geometry-preserving image retrieval algorithms in terms of accuracy