The set S of cells is called a seed set, as the goal is to seed the contour propagation algorithm defined above with cells from S. We describe a specialized algorithm for determining a seed set for a structured mesh, and cite performance characteristics for several subvolumes from the Visible Human data.
Initially, let the set S consist of all cells c, with an associated range T(c) which is the range of scalar values taken on over the given cell, R(c). We proceed in visiting each cell in a regular marching order, propagating responsibility ranges in the direction of the cell traversal. Intuitively, a cell must treat a responsiblity range as the set of values for which previously visited cells have deferred responsibilty. Upon visiting a cell, incoming responsibility must either be further propagated, or accounted for by the current cell. The end result is that a set of simple tests and range operators can be applied to each cell in a regular traversal, rapidly pruning cells from the seed set S.
For a more complete discussion and details of seed set construction, as well as a proof of seed set construction, see [5].