Automatic 3D reconstruction: An overview of the TRACER project.

The problem of reconstructing the three-dimensional structure of an object from a set of two-dimensional images arises in medicine, biology, and other fields. In neurobiology it is important to reconstruct the individual 3D structure of brain cells, or neurons, from sets of serial section electron micrographs. The TRACER project is an empirical attempt to combine appropriate low-level techniques with procedural domain knowledge to reconstruct individual, unbranching neuron sections in a particular set of input data. TRACER has achieved a certain amount of success in this subset of the reconstruction problem. This paper has two purposes. The first is to overview the problem, previous approaches to it, and describe the TRACER program as it evolved. The second purpose is to indicate fundamental limitations in the approach taken by the TRACER project, and indicate the most important areas for future work.