Adaptive reconstruction of freeform objects with 3D SOM neural network grids

There are several open problems that are viewed as a bottleneck in the reverse engineering process: (1) The topology is unknown; therefore, point connectivity relations are undefined. (2) The fitted surface must satisfy global and local shape preservation criteria, which are undefined explicitly. The reconstruction is based on parameterization and fitting stages. However, the above problems are influenced mainly by the parameterization. To overcome the above problems, the neural network self-organizing map (SOM) method is proposed for creating a 3D parametric grid. The main advantage of the SOM method is that it detects the orientation of the grid and the position of the sub-boundaries. Then through an adaptive process the neural network grid is converged to the sampled shape. The SOM method is applied directly on a 3D grid and avoids projection anomalies, which are common to other methods. For the surface fitting stage the random surface error correction fitting method, which is based on the SOM method, was developed and implemented. (C) 2002 Elsevier Science Ltd. All rights reserved.