Object Separation using Random-Walk Distance Map
Recipe realizing an object separation workflow relying on Random-Walk Distance Map. It is more robust to noisy segmentation or non-spherical shapes, compared to Separate Object.
Object Separation workflows typically rely on a Euclidean or Chamfer Distance map for the detection of local extrema to identify the center of individual objects, and a watershed transform to identify separating lines. In Amira-Avizo-PerGeos, the module Separate Object implements such a workflow.
However, such workflows are producing best results on clean segmentation of spherical objects, but tend to yield over and/or under-splitting when the segmentation is noisy and when the shape of objects becomes elongated, flat, or non-convex.
Compared to traditional distance maps, the Random-Walk Distance Map is less sensitive to noise in the segmentation, or to the shape factor of the objects, and provides more robust results for object separation.
This Xtra encapsulate the workflow of separating a binary segmentation into individual objects in a single, simple to use recipe.
Data courtesy: Mason Dean & Ronald Seidel (MPI Postdam-Golm), David Knoetel (Zuse Institute Berlin)