How to localize 3D-views in space?

A full 360° virtual 3D-view of an object is commonly put together from many different overlapping 3D-measurements. The task of putting this puzzle automatically together (“coarse registration”), is still a great challenge - although a child can do it. How does our brain manage the coarse registration? Obviously, we look for areas or points with big “saliency”, points that are eye-catching. We introduce an information theoretical concept, to find and mark those “eye-catching” points. The concept uses “pragmatic information”, which tells us, how useful the information for the specific task is. This pragmatic information allows a quantitative judgement about the value of different eye-catching points, for the registration of the data. The concept is implemented and we demonstrate its ability to reliably register 3D-views of a wide range of objects, within a few seconds.