Fast adaptive selection of best views

Abstract

Automatic computation of best views of objects is very useful. For example, they can be used as the starting point of a scene exploration, or to enrich galleries of objects available through Internet by adding an image a model that may help to decide if it is worth downloading. To select the most interesting viewpoint of an object, we use the so-called viewpoint entropy. The best view is the one which gives the most information of the object being inspected. In this paper we present an adaptive method to compute best views. Our adaptive scheme allows to improve over previous approaches the time of the selection of best views by an order of magnitude, and achieve a nearly interactive rate.