I have an image which is a result of active contour segmentation that leaked from an organ of interest into another (there is a small contact area with no contrast between the objects). So, there are two objects - generally smooth and circular but with irregularities. The contact area is at most 1/3 the width of the smaller object. I tried: 1) Morphological operations - generally works to separate the objects but the size of the structuring element is large enough (circle of size 9 or so) that upon dilation, it does not return the same mask. 2) Watershed - gives horrible oversegmentation due to objects' irregular shapes. It seems I should be able to do something with the distance transform, though. I know that the smaller object has at least a certain size (basin depth). Also, the boundary between two actual objects is far shorter than that between any basin-like areas within either of the objects
Question: Can you suggest a way to apply this knowledge in a watershed-like segmentation? Would some other approach be preferable?
Thanks in advance, Bogdan