Viewpoint Selection by Feature Measurement on the Viewing Plane

This paper presents a new method for selecting good viewpoints to efficiently observe an object. It defines a view-dependent curvature to measure the distribution and appearance of the geometric features on the viewing plane, and based on the curvature, it computes an entropy value for every candidate viewpoint. Afterwards, the viewpoints with the maximal entropy values are selected as the good viewpoints. The results of experiments show that the viewpoints selected by the new method could have salient features visible as much as possible, very likely to achieve what is required by people to view an object. In comparison with existing methods, the new method is simple, without semantic computation, and works very well.