Scan Integration as a Labeling Problem

Integration is a crucial step in the reconstruction of complete 3D surface
model from multiple scans. Ever-present registration errors and scanning
noise make integration a nontrivial problem. In this paper, we propose a
novel method for multi-view scan integration where we solve it as a labeling
problem. Unlike previous methods, which have been based on various
merging schemes, our labeling-based method is essentially a selection strategy.
The overall surface model is composed of surface patches from selected
input scans. We formulate the labeling via a higher-order Markov Random
Field (MRF) which assigns a label representing an index of some input scan
to every point in a base surface. Using a higher-order MRF allows us to
more effectively capture spatial relations between 3D points. We employ belief
propagation to infer this labeling and experimentally demonstrate that
this integration approach provides significantly improved integration via both
qualitative and quantitative comparisons.