We present a method for low-level image segmentation. The problem is formulated in terms of the partition induced by a set of sources and a determined energy. We study two types of energies that process the image at different levels of analysis. The first class is based on the theory of minimal paths and produces a présegmentation of the image. In the second part, where the energy is defined through a region merging process, the simplified image is used to create hierarchies of partitions.