Extending the Differential Image Foresting Transform to Root-Based Path-Cost Functions with Application to Superpixel Segmentation

The Image Foresting Transform (IFT) is a general framework to develop image processing tools for a variety of tasks such as image segmentation, boundary tracking, morphological filters, pixel clustering, among others. The Differential Image Foresting Transform (DIFT) comes in handy for scenarios where multiple iterations of IFT over the same image with small modifications on the input parameters are expected, reducing the processing complexity from linear to sublinear with respect to the number of pixels. In this paper, we propose an enhanced variant of the DIFT algorithm that avoids inconsistencies, when the connectivity function is not monotonically incremental. Our algorithm works with the classical and non-classifical connectivity functions based on root position. Experiments were conducted on a superpixel task, showing a significant improvement to a state-of-the-art method.