Unsupervised and Supervised Image Segmentation Using Graph Partitioning

Abstract

Image segmentation is an important research area in computer vision and its applications in different disciplines, such as medicine, are of great importance. It is often one of the very first steps of computer vision or pattern recognition methods. This is because segmentation helps to locate objects and boundaries into images. The objective of segmenting an image is to partition it into disjoint and homogeneous sets of pixels. When segmenting an image it is natural to try to use graph partitioning, because segmentation and partitioning share the same high-level objective, to partition a set into disjoints subsets. However, when using graph partitioning for segmenting an image, several big questions remain: What is the best way to convert an image into a graph? Or to convert image segmentation objectives into graph partitioning objectives (not to mention what are image segmentation objectives)? What are the best graph partitioning methods and algorithms for segmenting an image? In this chapter, the author tries to answer these questions, both for unsupervised and supervised image segmentation approach, by presenting methods and algorithms and by comparing them.