Watershed (image processing)

The watershed transform (WST; . Engl Watershed Transformation) is a method for segmentation within image processing.

Principle

The method is applied to gray level images. The gray value is best interpreted as height information. With the gradual flooding of the gray value mountain watersheds will be built between adjacent reservoirs. In general this results in an over- segmentation of the image, especially in noisy images, for example in medical CT data. The image must either be pre-processed or regions must then be summarized in a merge step based on a similarity criterion. Alternatively can be used depending on the application, a variant of the WST.

The gray-scale image needed are obtained for example by calculating the gradient of the original image; watersheds should be erected here along strong edges later. For binary data, one can calculate the inverse of the Euclidean distance transform. The WST is then separated, for example related objects.

Variants

In the pre -filled WST only reservoirs are flooded that exceed a certain size.

In the hierarchical WST, the result is converted to a graph representation ( ie, the neighborhood relationships of the segmented regions are detected ) and it further WST be performed recursively. Problem: The watersheds are becoming ever wider.

WST based on the marker, the marker positions determined by flooding, is carried out, the user has set in advance or interactively obtained using morphological operators previously.

With interactive versions of the WST, it is possible to set or to create artificial watersheds to improve the segmentation result so-called include and exclude points.