Structure-oriented Networks of Shape Collections

摘要

We introduce a co-analysis technique designed for correspondencerninference within large shape collections. Such collections are naturallyrnrich in variation, adding ambiguity to the notoriously difficultrnproblem of correspondence computation. We leverage the robustnessrnof correspondences between similar shapes to address the difficultiesrnassociated with this problem. In our approach, pairs of similarrnshapes are extracted from the collection, analyzed and matchedrnin an efficient and reliable manner, culminating in the constructionrnof a network of correspondences that connects the entire collection.rnThe correspondence between any pair of shapes then amounts to arnsimple propagation along the minimax path between the two shapesrnin the network. At the heart of our approach is the introduction of arnrobust, structure-oriented shape matching method. Leveraging thernidea of projective analysis, we partition 2D projections of a shapernto obtain a set of 1D ordered regions, which are both simple andrnefficient to match. We lift the matched projections back to the 3Drndomain to obtain a pairwise shape correspondence. The emphasisrngiven to structural compatibility is a central tool in estimating thernreliability and completeness of a computed correspondence, uncoveringrnany non-negligible semantic discrepancies that may exist betweenrnshapes. These detected differences are a deciding factor inrnthe establishment of a network aiming to capture local similarities.rnWe demonstrate that the combination of the presented observationsrninto a co-analysis method allows us to establish reliable correspondencesrnamong shapes within large collections.