Abstract

This paper proposes a new multi-scale energy minimization algorithm which can be used to efficiently solve large scale labelling problems in computer vision. The basic modus operandi of any multi-scale method involves the construction of a smaller problem which can be solved efficiently. The solution of this problem is used to obtain a partial labelling of the original energy function, which in turn allows us to minimize it by solving its (much smaller)
projection. We propose the use of new techniques for both the construction of the smaller problem, and the extraction of a partial solution. We demonstrate our method on the problem of interactive image segmentation. Traditional multi-scale approaches for segmentation extract a partial solution using an image band around the boundaries of
the object segmentation obtained by minimizing the smaller problem. This strategy fails on objects with fine structures and complex topologies. In contrast, our novel approach
uses a min-marginal based uncertainty measure which allows us to handle such objects. Experiments show that our techniques result in solutions with low pixel labelling error.
Furthermore, they take the same or less amount of computation compared to traditional multi-scale techniques.