A Neural Network Model for Objects with Freeform Surfaces

Abstract

A two-layer neural network is proposed for modeling objects with freeform surfaces in computer graphics and computer vision. The parametric form describes not only geometric shapes and normal vectors but also texture and other optical features of arbitrary topology. It also facilitates animation. The implicit form models any convex surfaces and can be used for ray tracing. The algorithm is “neuronal” and suitable for parallel and distributed implementation. Using a well-trained network, the model can be used in computer-aided geometric design and model-based vision.