Unsupervised progressive parsing of Poisson fields using minimum description length criteria

This paper describes novel methods for estimating piecewise homogeneous Poisson fields based on minimum description length (MDL) criteria. By adopting a coding-theoretic approach, our methods are able to adapt to the the observed field in an unsupervised manner. We present a parsing scheme based on fixed multiscale trees (binary, for 1D, quad, for 2D) and an adaptive recursive partioning algorithm, both guided by MDL criteria. Experiments show that the recursive scheme outperforms the fixed tree approaches.