Foundations

Segmentation & edges

Morphology

Technique

Intermediate

Opening and Closing

Compose erosion and dilation to remove bright specks or fill small holes without manually editing the mask.

Morphology is nested under Segmentation & edges , so the broader pipeline usually still applies here.

graphics

image-processing

morphology

mask

webgpu

Interactive playground

Tweak the operator or scene live so the article connects to an immediate visual result.

Opening / closing

Chain morphology passes to remove specks or fill holes without manually touching every pixel.

Source image

Procedural input

WebGPU result

Composite output

Controls

Adjust the neighborhood size, repeat count, or decision threshold to see how the shader compounds the result.

Neighborhood radius: 2 Composite operator

Checking WebGPU support...

Family

Foundations -> Segmentation & edges -> Morphology

Binary-mask operators such as dilation, erosion, opening, and closing.

Builds on

4 topics

Read these first if you want the surrounding pipeline context.

Unlocks

0 next topics

Use these follow-ups when you want to keep turning the image-processing pipeline forward.

Learning paths

This topic appears in curated graphics progressions so the next step is obvious.

Choose this over that

Opening and closing are deliberate two-step repairs

Both operators combine erosion and dilation, but the order decides whether you remove bright specks or fill small dark holes.

Composite operator

Opening = erosion then dilation

Choose this when: Small bright specks should disappear while larger shapes stay mostly intact.

Choose something else when: The bigger problem is small dark gaps or holes inside the foreground.

Composite operator

Closing = dilation then erosion

Choose this when: Small holes and gaps should be sealed without permanently thickening the whole shape.

Choose something else when: You are trying to remove isolated bright noise instead of fill gaps.

Filtering & convolution Open topic

Gaussian Blur

Choose this when: The source is still a continuous image and should be smoothed, not treated as a binary mask.

Choose something else when: The mask already exists and needs structural cleanup.

Problem

Single-step dilation and erosion are useful, but often too blunt.

Typical mask defects come in two common forms:

isolated bright specks that should disappear
small dark gaps or holes that should be filled

You need a controlled two-step repair, not just unconditional growth or shrinkage.

Intuition

Opening and closing use the same two primitive operators:

erosion
dilation

The entire behavior is decided by the order.

Opening

opening = erosion -> dilation

This removes small bright foreground noise first, then regrows the surviving larger structures.

Closing

closing = dilation -> erosion

This fills small dark holes or narrow breaks first, then shrinks the result back toward its original boundary.

That order difference is the whole story.

Worked example

Opening

Suppose a mask contains one large object plus tiny bright specks.

Erosion removes the specks entirely and shrinks the large object.
Dilation regrows the large object.
The specks do not come back because they were erased completely in the first step.

Closing

Suppose a mask contains a mostly solid region with tiny holes.

Dilation grows the foreground so holes and narrow gaps seal.
Erosion trims the outside boundary back down.
The previously sealed holes stay filled if they were smaller than the structuring element.

Why opening and closing are not opposites

They use the same two primitives, but they are not interchangeable.

Opening is biased toward removing small bright features
Closing is biased toward filling small dark features

That makes them perfect “choose this over that” topics.

Complexity

If one morphology pass costs $O(WHk^2)$ , then opening or closing costs:

O(2WHk^2)

which is still just a constant-factor multiple of the primitive operation.

GPU implementation note

On WebGPU, opening and closing are naturally expressed as two passes:

render the first morphology result into an intermediate texture
run the second morphology pass from that intermediate texture into the final target

That makes them a good example of a small multi-pass graphics pipeline rather than a single operator.

When to choose each

Choose opening when

the mask contains isolated bright specks
thin false positives should disappear
the foreground should become cleaner, not more connected

Choose closing when

the mask contains tiny holes
boundaries have narrow cracks
nearly connected regions should fuse

Key takeaways

Opening and closing are two-pass morphology techniques
The order determines the effect
Opening removes small bright noise
Closing fills small dark gaps and holes
They are more selective than applying dilation or erosion alone

Practice ideas

Build one noisy mask with bright specks and another with small holes, then compare opening and closing on both
Vary the morphology radius and note the largest defects each operator can still fix
Implement opening and closing as two WebGPU passes with an intermediate texture

Relation to other topics

Dilation and erosion are the primitive operators opening and closing are built from
Thresholding often produces the mask that later needs opening or closing
Gaussian blur is a continuous-image cleanup step from earlier in the pipeline, not a binary-mask repair tool

Build on these first

These topics supply the mental model or preceding stage that this page assumes.

Foundations

Concept

Image Processing Fundamentals

Build the mental model behind blur, thresholding, edges, and morphology: pixels, neighborhoods, kernels, masks, and why local operators compose into pipelines.

Segmentation & edges

Algorithm