The H-Minima Transform was invented to reduce the over-segmentation in watershed algorithms by filtering out noise in the darker regions of gray value images.

The method searches for local minima in the gray value image and suppresses them if their depth is less than or equal to the Minimum Height. The depth of a local minimum is defined as the difference between the gray value of the local minimum and the next local maximum gray value.

The simplified 2D example below shows neighboring gray values in gray bars. The corresponding depth of each local minimum is visualized in blue. For comparison, the given Minimum Height is visualized as a red rectangle. Note that the depth of the two minima in the center is shallower than the Minimum Height, while the depth of the other two is deeper than the Minimum Height. Thus, as shown below, the two in the middle will be suppressed, while the other two will not.

Tolerance represents the relative change in image mean gray values. If 0 is entered for Tolerance, the runtime may be very long as it tries to find a perfectly stable solution. However, in most cases, a slightly higher tolerance, the default of 0.001 for example, already leads to good results.

Click Apply to apply the filter on the gray value image. The following example has already been filtered with the Non-Local Means Filter and a Morphological Gradient. Then, the H-Minima Transform was applied with a minimum height of 30.