Multi-faceted Image Retrieval by Concept Lattice Querying and Navigation

Typical content-based image retrieval (CBIR for short) methods aim at capturing image similarity by relying on some specific characteristic of images such as color, texture and shape. Existing CBIR systems use these methods in order to support users in accessing the knowledge embedded in images. However, these systems allow to explore the image space in just one way, in the sense that it is not possible to navigate the space by simultaneously applying multiple similarity criteria. The model we propose addresses this problem by representing the search for the images similar to a given image as the exploration of a lattice of (non-disjoint) image clusters, induced by a natural ordering criterion, based on similarity measures. The exploration proceeds in one of two basic ways: (1) by querying, the user can jump to any cluster of the lattice, by specifying the criteria that the sought cluster must satisfy; or (2) by navigation: from any cluster, the user can move to a neighbor cluster, thus exploiting the ordering amongst clusters.