We study the scenario of graph-based clustering algorithms such as spectral clustering. Given a set of data points, one first has to construct a graph on the data points and then apply a graph clustering algorithm to find a suitable partition of the graph. Our main question is if and how the construction of the graph (choice of the graph, choice of parameters, choice of weights) influences the outcome of the final clustering result. To this end we study the convergence of cluster quality measures such as the normalized cut or the Cheeger cut on various kinds of random geometric graphs as the sample size tends to infinity. It turns out that the limit values of the same objective function are systematically different on different types of graphs. This implies that clustering results systematically depend on the graph and can be very different for different types of graph. We provide examples to illustrate the implications on spectral clustering.

When biological specimens are cut into physical sections for three-dimensional (3D) imaging by confocal laser scanning microscopy, the slices may get distorted or ruptured. For subsequent 3D reconstruction, images from different physical sections need to be spatially aligned by optimization of a function composed of a data fidelity term evaluating similarity between the reference and target images, and a regularization term enforcing transformation smoothness. A regularization term evaluating the total variation (TV), which enables the registration algorithm to account for discontinuities in slice deformation (ruptures), while enforcing smoothness on continuously deformed regions, was proposed previously. The function with TV regularization was optimized using a graph-cut (GC) based iterative solution. However, GC may generate visible registration artifacts, which impair the 3D reconstruction. We present an alternative, multilabel TV optimization algorithm, which in the examined samples prevents the artifacts produced by GC. The algorithm is slower than GC but can be sped up several times when implemented in a multiprocessor computing environment. For image pairs with uneven brightness distribution, we introduce a reformulation of the TV-based registration, in which intensity-based data terms are replaced by comparison of salient features in the reference and target images quantified by local image entropies.