VISIBLE SURFACE RECONSTRUCTION VIA LOCAL MINIMAX APPROXIMATION

Visible surface reconstruction is an important early stage in the processing and analysis of 3-D imagery, as it provides a compact and manipulable representation of the 3-D structure of perceived objects in a scene. We describe an algorithm for reconstructing visible surfaces from scattered surface depth/orientation data using approximation methods. The basic unit of the proposed algorithm is a minimax approximation routine which operates over regions of local surface support. The minimax approach is attractive for several reasons and the computational cost is not high using fast algorithms. A window-based scheme is used to achieve a smooth surface reconstruction while maintaining discontinuities arising from object occlusions or sudden changes in surface orientation; the discontinuities are detected as part of the algorithm. Results of the technique as applied to several test objects are provided.