In this paper we propose a novel approach to parallel image segmentation of volume images using the watershed transformation on stream computing platforms. The watershed transformation is a powerful mathematical morphology method for gray-scale image segmentation. It is widely used in medical, technical, biological and other image analysis applications, mostly for extracting homogeneous areas with respect to the gray-value gradient. However the watershed-transformation is a very computation intensive task. With the increasing programmability of graphic processing units, a cheap and powerful high performance computing platform is available. We present an algorithm that was especially adapted for stream processing and give a brief formal explanation of the correctness of our method. We also discuss our exemplary implementation for the NVidia CUDA platform and show speedup measurements for sample volume datasets.