Automatic marker detection for blob images

The direct application of the Watershed Transform to images with moderate contrast between foreground and background often produces segmentation where the segmented areas will still have a lot of background included. One typical example of this is the segmentation of blob images. To narrow the border of these areas, the Watershed Transform may be applied to the gradient magnitude of the image, but it results in over-segmentation. Marker Controlled Watershed Transform is one of the best approaches for the segmentation of this kind of images, however, in most cases it is impracticable to manually select the markers. This paper presents a technique that allows the automatic detection of markers on blob images and subsequent segmentation.