This paper describes a technique for enhancing certain features in grayscale images. Of particular interest are the class of objects that are reasonably large in extent, but are only faintly darker or lighter than the background. An example of such objects is the mandibular canal which appears in panoramic dental X-ray images. Identification of this canal is required in some dental and orthodontal investigations. Traditional image segmentation techniques often fail to detect the full extent of the canal due to the large amount of structural noise in the image. We propose a new method for the enhancement of this class of objects and the subsequent segmentation task. The flow of a fluid is simulated over the image topology, allowing fluid to settle in local minima and, by application of a difference image, enhancing the visibility of features that are characterized by a significant spatially-distributed local minimum. The procedure is similar to the watershed algorithm in concept, visualizing the movement of fluid over the image surface to draw conclusions about significant local minima. Our approach is different however since it is not aimed at segmenting the image, but enhancing distributed local minima. We consider the flow of fluid from high lying to lower lying areas under gravity. This is analogous to the rain fall method of filling the catchment basins in watershed segmentation. Various models of flow, based on co- operative networks are presented and discussed. Post processing is applied to reduce the amount of false outputs. We demonstrate that our proposed method is more suitable than simple edge detection or the watershed algorithms for the enhancement and segmentation of the mandibular canal.