Classical Watershed 3D Algorithm

## Pre-filters to apply before this algorithm

- Noise reduction filters (like median, top hat)
- Laplacian of Gaussian that can be found in Misc Filters 3D gives very good results

## The Algorithm

### Description:

- The watershed is seeded with Regional Minima of the watershed map (given as a parameter, which is usually the gradient of the input image), in order to limit over-segmentation
- Propagation : propagation is done in the ascending order of the watershed map, starting from the seeds. It is a 6-connectivity propagation. Propagation is done on the whole image.
- As there is no constraint on seed and no limit for propagation, this algorithm detects well the edges of objects.

### Over-segmentation

This algorithm can produce over-segmentation. This can be corrected with (at least) 3 operations:- Reducing the noise during before computing the watershed map
- Increasing the derivation radius and/or applying filters on the watershed map
- Using the post-filter Merge Regions

## Post-filters

- As the algorithm segments the whole image into regions, regions corresponding to background have to be removed. This can be done using the post-filter Erase Spots. It can be applied several time in order to have several criteria
- In case of over-segmentation: Merge Regions
- Classical morphological operations

## Usage:

- Spots-like objects like chromocenters
- If objects are too small, they might not be segmented properly.