This work constitutes a first approach on image segmentation based on the recently proposed morphological scale-space theory. We introduce an idempotent smoot hing operation, in the corresponding scale-space, and analyze some of its main features concerning the monotonicity of the image extrema and the way these extrema merge in a multiscale simplification process. We also define some basic criteria to control the merging of the image extrema across scales to obtain good markers for segmentation. As we will illustrate, these methods take into account only local information of the image and yield sound segmentation results, mainly in those applications where the regions to be segmented can be characterized (marked) by the extrema of the image function.