The use of multiple machines to reconstruct large 3-D volumes from 2-D images

The use of higher level pattern recognition for 3-D reconstruction of objects from multiple 2-D images can be computationally very intensive. In our applications, we needed to process data sets consisting of up to 150 cross-sectional images and automatically reconstruct the surfaces describing the object. On a Silicon Graphics Indigo workstation, such a task normally takes one hour. By splitting up the reconstruction among many such machines using high level Unix calls, the job can be handled effectively with a corresponding decrease in reconstruction time, without significantly affecting the performance of any one machine.