Some Experiments in Applying Inductive Inference Principles to Surface Reconstruction

This paper presents an application of the Minimum Description-Length (MDL) principle of inductive learning to the surface reconstruction problem of computer vision. The application demonstrates that the M D L principle can be applied to practical problems while preserving its convergence properties. It also illustrates how at least one aspect of computer vision (i.e., surface reconstruction) can be treated as an inductive inference problem using this principle. This has the advantage that the convergence properties of the M D L principle enable an exact surface reconstruction to be obtained asymptotically as the number of available data points increases. Moreover, convergence to the true surface occurs independent of the amount of error in the measurements.