Dilation

Problem

A binary mask is often almost correct but not quite:

• tiny gaps split a region in two
• lines are thinner than they should be
• nearby foreground pieces fail to touch

You want to grow the foreground outward in a controlled way.

Intuition

Dilation asks:

Does this pixel have any foreground neighbor inside the structuring element?

If the answer is yes, the output becomes foreground.

So dilation behaves like a local max over a binary mask.

The operator

Using a 3 x 3 structuring element, each output pixel becomes 1 if any pixel in that neighborhood is 1.

In pseudocode:

output[x, y] = 1 if any input neighbor in the window is 1

That causes bright regions to expand outward.

Worked example

Suppose a 1D mask is:

0 0 1 0 0

With a radius-1 dilation, the foreground grows:

0 1 1 1 0

In 2D, the same effect thickens shapes and can bridge narrow gaps.

What dilation is good at

• reconnecting nearly-touching regions
• thickening thin lines
• filling tiny breaks
• making later selection logic less fragile

But dilation also has a cost: it can overgrow objects and blur their binary boundaries outward.

Complexity

For an image with $W \times H$ pixels and a structuring element of size $k \times k$:

• direct dilation costs $O(WHk^2)$

For small fixed kernels, that is often perfectly practical.

GPU implementation note

Dilation maps cleanly to WebGPU:

• each output pixel checks a local neighborhood
• there are no cross-pixel dependencies beyond that neighborhood
• the operator is effectively a local max on binary values

That makes it easy to animate the effect of increasing the morphology radius.

When to choose it

Choose dilation when:

• foreground regions should expand
• tiny gaps should seal
• thin structures should be reinforced

Choose something else when:

• the mask is already too thick
• small bright noise should disappear rather than spread
• you need a more selective two-step cleanup

Key takeaways

• Dilation grows foreground regions
• It is the binary-mask counterpart of a local max
• It is excellent for closing tiny breaks, but it can also overgrow shapes
• Dilation is one of the two primitive morphology operators
• Composite tools such as closing are built from it

Practice ideas

• Dilate a sparse mask with radius 1, 2, and 3 and note when separate components merge
• Compare direct dilation with closing on a mask that has both narrow gaps and isolated noise
• Build a threshold -> dilation pipeline and describe what gets fixed and what gets worse

Relation to other topics

• Thresholding often produces the mask that dilation then repairs
• Erosion is the opposite primitive: it shrinks foreground rather than grows it
• Opening and closing combine dilation and erosion to solve more targeted cleanup problems