Scan integration as a labelling problem

Abstract

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.

Density adaptive trilateral scan integration method

“…Furthermore, as with other clustering-based integration methods [24][25][26][27][28], the new method has high integration efficiency because the redundancies in overlapping areas can be eliminated completely. The new method is also simple and fast, since the input and output are both point set surfaces, and the trilateral shifting procedure and mean-shift clustering algorithm are easy to implement.…”

“…The clustering approaches based integration method fuses scans/MRIs by clustering homogeneous points in overlapping areas and, thus, weighted averaging the corresponding points to remove redundancies and noises [24][25][26][27][28]. Zhou and Liu [24] first proposed a k-means clustering scheme to integrate scans/ MRIs.…”

“…The fuzzy c-means clustering approach was employed to fuse overlapping points in the nonfeatured areas, and the k-means clustering approach was employed to fuse overlapping points in the featured areas. Recently, the authors proposed a labeling-based method [27]. By introducing a higher-order Markov random field to formulate the labeling, the spatial relations between overlapping points were captured more effectively.…”

“…The previously reported overlapping-area-detection algorithms can be divided into two categories: 3D-grid-based methods [5][6][7][8] and singlethreshold-based methods [24][25][26][27][28]. The former type of method divides the nodes into a series of 3D cubes, and the points, coming from different scans/MRIs, falling into the same cube are supposed to be overlapping points [5][6][7][8].…”

“…The single-threshold-based methods [24][25][26][27][28] employ a fixed threshold to split overlapping and nonoverlapping areas. This type of method can correctly separate overlapping areas in most situations.…”