Ask HN: How to handle images resistant to edge detection?

What tools or techniques have you found useful to detect objects, that are resistant to edge detection, during image processing for computer vision workflows?

Where one object bleeds into the other gradually.

Two examples of this include:

1. Close image of Foot prints in beach sand - edge detection focus on the edges of individual grains of sand rather than the gradient edge of the track.

For an example see: https://github.com/TrackerLounge/TrackingAndComputerVision - second image.

2. A halibut (fish) half in and half out of plankton-rich partially reflective water. The color of the fish and water can be so close that it becomes very difficult to programmatically detect where one ends and the other begins.

3. Cryptic sea creatures (e.g. mussels, clam, flatfish, etc.) half-buried in mud on sea floor.

I run into this question in image processing everywhere I look.

In these "gradient edge" scenarios what have you found to work best?