Determining the viewing points for 3 - D objects in a dynamic retrieval system

The dynamic retrieval system is considered as a means to construct a database for actual three-dimensional (3—D) objects, such as commercial goods, art works, animals and plants. In the dynamic retrieval system, the image data are given as the input continuously along the virtual semisphere covering the object. Since a large amount of image data is produced in this way, only N representative image data are extracted and stored. Apart from those representative image data, dynamical image data are prepared for rotation and display of the object in the partial area of the terminal screen. The user rotates the object displayed on the partial screen in an interactive way using a joy stick or other device. The outline of the object is examined and the next viewpoint to observe the object is determined. Then the representative image data corresponding to that viewpoint is displayed on the terminal. A problem in this dynamic retrieval is how to select the N representative image data from the continuously observed image data. For this problem, this paper proposes the following algorithm. Two images are taken by viewing the object from two different viewpoints. The variety of object sampling points is defined as an objective measure to represent the difference of information as sensed substantially by the human observer. Then, using that variety of object sampling points, the viewpoint for the representative image data is determined. A result of experiment is presented where the proposed method is applied to determine the representative viewpoint.