A combined algorithm for image segmentation using neural networks and 3D surface reconstruction using dynamic meshes.

Authors

  • C. C. Reyes Aldasoro Departamento de Sistemas Digitales Instituto Tecnológico Autónomo de México. Río Hondo No. 1, Tizapán San Angel, 01000 México D.F
  • M. E. Algorri Departamento de Sistemas Digitales Instituto Tecnológico Autónomo de México. Río Hondo No. 1, Tizapán San Angel, 01000 México D.F

Abstract

 

Reconstructing the surface from a set of unstructured points to build a 3D model is a problem that arises in many scientific and industrial fields as new 3D scanning technology is able to produce large databases of full 3D information. 3D surface reconstruction is also important after segmenting sets of 2D images to visualize the 3D surface represented by the segmentation. In this paper, we propose an algorithmic methodology that obtains a series of segmentation of human head tomographies, produces a set of unstructured points in the 3D space, and then automatically produces a surface from the set of unstructured 3D points about which we have no topological knowledge. The methodology can be divided into two stages. First, tomographic images are segmented with a Neural Network algorithm based on Kohonen.s Self-Organising Maps (SOM). The output neurons that have adapted to the image, are a series of 3D points that will be fed to the second stage. Next, our method uses a spatial decomposition and surface tracking algorithm to produce a rough approximation S. of the unknown manifold S. The produced surface S. serves as initialization for a dynamic mesh model that yields the details of S to improve the quality of the reconstruction.

 

Downloads

Download data is not yet available.

Downloads

Published

2000-08-28

How to Cite

Reyes Aldasoro, C. C., & Algorri, M. E. (2000). A combined algorithm for image segmentation using neural networks and 3D surface reconstruction using dynamic meshes. Revista Mexicana De Ingenieria Biomedica, 21(3), 9. Retrieved from http://rmib.mx/index.php/rmib/article/view/402

Issue

Section

Research Articles

Dimensions Citation