matching against one image takes an average of 100 ms. It clearly emerges how reducing the number of database entries can greatly decrease the computation effort. In fact, if no localization information is used, the time required by the matching component increases linearly with the number of artworks in the museum. As Fig. 7 shows, assuming an average of 5 artworks per room and a total of 10 rooms, exploiting localization results in an average processing time per frame of 1700 ms. On the other hand, matching against the entire museum database results in an average of 6200 ms (a 365% increase in required computational times).