Fast watershed transform is a simulation of the immersion analogy in
a descrete space. It starts with sorting image pixels according to their
intensity levels and store them in their corresponding FIFO structures.
Beginning with the FIFO containing pixel coordinates with the lowest intensitiy,
a queue-based simulation of immersion. As described in detail previously,*
catchment basins at an intensity
level h and the pixels with the next intensity level h+1
may have three kinds of relationship (left image).
1) More than one catchment basin (1,2, and 3) are included in an isolated
region of Th+1.
2) A catchment basin (4) is included in another isolated region of
3) No catchment basin is included.
At h, water emerging from catchment basins 1, 2, and 3 will
meet at SKIZ (skeleton by influence zones) depicted
as dark lines at the right image. Water from catchment basin 4 will
fill the isolated region. Finally, the isolated region labeled 5 can be
detected as a new minimum. This relation holds for all intensity
levels higher than h.
* For complete algorithm, please read L. Vincent and P. Soille, "Watersheds
in digital spaces: an efficient algorithm based on immersion simulations,"
IEEE Trans. Pattern Anal. Machine Intell., 13(6): 583-598, June 1991