2D image segmentation using minimum spanning trees

This paper presents a new algorithm for partitioning a gray-level image into connected homogeneous regions, which reduces the computational complexity of the region partitioning problem by constructing a minimum spanning tree representation of agray- level image by minimizing the sum of variations of gray levels over all subtrees.

Abstract

This paper presents a new algorithm for partitioning a gray-level image into connected homogeneous regions. The novelty of this algorithm lies in the fact that, by constructing a minimum spanning tree representation of a gray-level image, it reduces a region partitioning problem to a minimum spanning tree partitioning problem, and hence reduces the computational complexity of the region partitioning problem. The tree-partitioning algorithm, in essence, partitions a minimum spanning tree into subtrees, representing different homogeneous regions, by minimizing the sum of variations of gray levels over all subtrees under the constraints that each subtree should have at least a specified number of nodes, and two adjacent subtrees should have significantly different average gray-levels. Two (faster) heuristic implementations are also given for large-scale region partitioning problems. Test results have shown that the segmentation results are satisfactory and insensitive to noise.