Performance Evaluation of Self-Organising Map Model in Organising the Unstructured Data

Abstract

Surface reconstruction becomes a difficult task in reverse engineering when the data obtained during the data acquisition process is unstructured. The unstructured data do not contain the connectivity information required to represent the surface correctly with the least error. Hence, it should be organised to obtain the connectivity information. Various types of Self-Organising Map (SOM) models are utilised in the previous works to organise the unstructured data and represent the surface. However, the performance of the SOM models is affected when different topologies are involved in the organising process. Therefore, the purposes of this experiment are to evaluate the performance of the SOM models with different topologies and to determine the limitation of the various SOM models. The SOM models involved are 2-D SOM, 3-D SOM, Cube Kohonen (CK) SOM, and Spherical SOM (SSOM). Three 3-D unstructured closed surface data sets are applied in this experiment to evaluate the models. The experimental results show that the CKSOM and SSOM models can represent the closed surface correctly with a medium speed. Overall, the CKSOM model performs better than the SSOM model as its grid size can be tuned and it achieved 9 out of 9 minimum error in presenting the surface.