Using self-creating neural network for surface reconstruction

Surface reconstruction is a very important step in surface rendering of medical virtual reality. In addition to conventional methods, many researchers have employed growing cell structures (GCS) neural networks to implement surface reconstruction. Due to its characteristic of learning vector quantization (VQ) using GCS in surface reconstruction could lead to some serious problems. To solve these problems, we use a hybrid network that incorporates GCS and BNN to perform surface reconstruction. The method is adaptive, in the sense that the regions of high curvature will be represented with more and smaller polygons, and the rest with less and bigger polygons. The excellent topological preserving capability of GCS allows us to use the curvature of topological mapping to replace the curvature of original input data. Simulation results have shown that the proposed hybrid network can achieve better reconstruction result than does the GCS network