Image Segmentation by Image Foresting Transform with Non-smooth Connectivity Functions

In the framework of the Image Foresting Transform (IFT), there is a class of connectivity functions that were vaguely explored, which corresponds to the non-smooth connectivity functions (NSCF). These functions are more adaptive to cope with the problems of field in homogeneity, which are common in MR images of 3 Tesla. In this work, we investigate the NSCF from the standpoint of theoretical and experimental aspects. We formally classify several non-smooth functions according to a proposed diagram representation. Then, we investigate some theoretical properties for some specific regions of the diagram. Our analysis reveals that many NSCFs are, in fact, the result of a sequence of optimizations, each of them involving a maximal set of elements, in a well-structured way. Our experimental results indicate that substantial improvements can be obtained by NSCFs in the 3D segmentation of MR images of 3 Tesla, when compared to smooth connectivity functions.