An investigation into graph cut parameter optimisation for image-fusion applications
Xiao Bao Clark, Jackson G. Finlay, Andrew J. Wilson, Keith L. J. Milburn, Minh Nguyen, Christof Lutteroth, Burkhard Claus Wünsche · 2012
The graph cut technique has been employed successfully in a large number of computer graphics and computer vision related problems. The algorithm has yielded particularly impressive results in the field of image fusion, e.g., texture tiling, image stitching, and image and video editing. An analysis of the literature shows that authors use different variations of the algorithm, such as different cost functions and parameters. However, there are no detailed investigations on how these parameters influence results and what parameters are most suitable for what type of application. In this paper we analyse the use of graph cut algorithms in different image fusion applications. We list and classify relevant parameters, suggest new cost functions for seam optimisation, and analyse the effect of parameter choices on different application scenarios. Based on the results we develop guidelines assisting users to employ the graph cut technique effectively in different image fusion applications.