The 2 pass raster segmenter is simple, fast and is often quoted in the literature. Unfortunately, it tends to oversegment images even in the presence of small amounts of noise. In this paper we present a generalization of this approach where we discover regions by taking multiple random paths through an image. This approach fares better but still over segments an image. Yet, an analysis of region density shows that the underlying image structure can be discovered from the path based segmentation. Indeed, the discovered edges are comparable to those discovered by the widely used mean shift algorithm.