An efficient method to map a regular mesh into a 3D neural network

In 3D computer vision a relevant problem is to match a "source" image dataset with a "target" image dataset. The matching problem can be faced using a neural net approach, where the nodes are related to the image voxels and the synapses to the voxel information. This paper presents an improvement of the "Volume-Matcher 3D" project, an approach for a data-driven comparison and registration of three-dimensional images based on 3D neural networks. The approach has been improved by introducing a method for an efficient mapping of a regular mesh into a 3D neural network in order to reduce the high computational complexity. The algorithms developed have been tested on real cases of interest in the field of medical imaging. The software has been implemented on a high performance PC using the AVS/Express^TM software package for volume reconstruction