The PIBE personalizable image browsing engine

In this paper we describe PIBE, a new Personalizable Image Browsing Engine that allows an effective visual exploration of large image collections combining computer vision and database techniques. The principal features of PIBE include the possibility of modifying the browsing structure by means of graphical personalization actions provided by the visual interface, and of persistently storing such customizations for subsequent browsing sections. The PIBE hierarchical browsing structure, called Browsing Tree, can be locally customized, thus avoiding global reorganizations, which are clearly unfeasible with large collections. Indeed, each node of the Browsing Tree has associated a cluster of images and a specific dissimilarity function. We present the basic principles of the PIBE engine, and report some experimental results showing the effectiveness and the efficiency of the browsing and personalization functionalities provided.