Avoiding the Global Sort: A Faster Contour Tree Algorithm

Abstract

We revisit the classical problem of computing the contour tree of a scalar field $$f:\mathbb {M}\rightarrow \mathbb {R}$$f:MźR, where $$\mathbb {M}$$M is a triangulation of a ball in $$\mathbb {R}^d$$Rd. The contour tree is a fundamental topological structure that tracks the evolution of level sets of f and has numerous applications in data analysis and visualization. All existing algorithms begin with a global sort of at least all critical values of f, which can require (roughly) $$\Omega (n\log n)$$Ω(nlogn) time, where n is the number of vertices of the mesh. Existing lower bounds show that there are pathological instances where this sort is required. We present the first algorithm whose time complexity depends on the contour tree structure, and avoids the global sort for non-pathological inputs. (We assume that the Morse degree of the function f at any point is at most 3.) If C denotes the set of critical points in $$\mathbb {M}$$M, the running time is roughly $$O(N+\sum _{v \in C} \log \ell _v)$$O(N+źvźCloglv), where $$\ell _v$$lv is the depth of v in the contour tree and N is the total complexity of $$\mathbb {M}$$M. This matches all existing upper bounds, but is a significant asymptotic improvement when the contour tree is short and fat. Specifically, our approach ensures that any comparison made is between nodes that are either adjacent in $$\mathbb {M}$$M or in the same descending path in the contour tree, allowing us to argue strong optimality properties of our algorithm. Our algorithm requires several novel ideas: partitioning $$\mathbb {M}$$M in well-behaved portions, a local growing procedure to iteratively build contour trees, and the use of heavy path decompositions for the time complexity analysis.