What counts as a 'stray' pixel?

Any non-transparent pixel whose 8 surrounding neighbors (the ring of pixels directly around it) include fewer than the chosen minimum number of other non-transparent pixels. Isolated single specks typically have 0 non-transparent neighbors and get removed by default.

How is this different from the edge softener or refiner?

The softener blurs the alpha channel and the refiner snaps extreme alpha values, but neither specifically targets isolated single-pixel noise the way this despeckle pass does. This tool looks at each pixel's local neighborhood, not just its own alpha value.

What does the minimum-neighbors setting control?

It's the threshold for how connected a pixel needs to be to survive. A low value (like 1) only removes fully isolated pixels; a higher value (like 4-5) is more aggressive and can also erode thin lines or small details.

Will this affect the interior of a solid shape?

No. Interior pixels of a solid, filled shape are surrounded by other non-transparent pixels and easily clear the minimum-neighbor threshold, so they're left untouched. Only sparse, isolated pixels near edges are affected.

PNG Edges Cleaner - Remove Stray Speck Pixels Online

Overview

Introduction

This tool cleans up stray, isolated pixels near a PNG's edges, entirely client-side, using a simple neighbor-counting despeckle pass.

It's useful for tidying up single-pixel artifacts left behind by a lossy background-removal or masking step.

What Is PNG Edges Cleaner?

A client-side PNG despeckling tool that examines each non-transparent pixel's 8 surrounding neighbors and removes (makes fully transparent) any pixel that doesn't have enough non-transparent neighbors to be considered part of a real shape.

It's a simple morphological cleanup pass, distinct from blurring (which softens everything) or thresholding (which only looks at a pixel's own alpha value).

How PNG Edges Cleaner Works

For each non-transparent pixel, the tool counts how many of its 8 surrounding pixels (up, down, left, right, and the four diagonals) are also non-transparent.

If that count is below the chosen minimum-neighbors setting, the pixel is treated as noise and its alpha is set to 0; otherwise it's left completely unchanged.

When To Use PNG Edges Cleaner

Use it after background removal or masking to clean up leftover single-pixel or few-pixel specks that weren't fully caught by the main operation.

It's also useful as a quick sanity pass before compositing a cutout onto a new background, where stray specks would otherwise be visible.

Often used alongside PNG Edges Refiner and PNG Background Remover.

Features

Advantages

Runs entirely client-side; the uploaded image is never sent to a server.
Targets isolated noise specifically, without blurring or thresholding the rest of the image.
Adjustable sensitivity via the minimum-neighbors setting.

Limitations

A high minimum-neighbors setting can also erode thin, legitimately-isolated details like fine hair strands or single-pixel-wide lines.
It only considers each pixel's immediate 8-neighbor ring, so small clusters of 2-3 noise pixels that reinforce each other's neighbor count may survive.

Examples

Remove a single stray speck

Input

A cutout PNG with one isolated opaque pixel floating in transparent space, minNeighbors 2

Output

That pixel becomes fully transparent; the main subject is untouched

An isolated pixel with 0 non-transparent neighbors falls below the threshold of 2.

Preserve a solid shape's edge

Input

A filled circle shape, minNeighbors 2

Output

The circle's edge and interior pixels remain fully intact

Edge pixels of a real shape have enough non-transparent neighbors along the boundary to survive.

Best Practices & Notes

Best Practices

Start with the default minimum-neighbors value of 2, which removes fully isolated specks without eroding real edges.
If your subject has fine, thin details (hair, whiskers, thin lines), use a lower minimum-neighbors value or skip this step for those images.

Developer Notes

cleanPngEdges is a pure lib function operating on a plain Uint8ClampedArray PixelBuffer: for each non-transparent pixel it counts non-transparent pixels among its 8-connected neighbors and zeroes the alpha of any pixel below the configured threshold — no DOM access until the component paints the result for preview/download.

PNG Edges Cleaner Use Cases

Cleaning up single-pixel artifacts left after background removal
Tidying a mask before compositing it onto a new background
Removing sensor-noise-style specks from a scanned or photographed cutout

Common Mistakes

Setting the minimum-neighbors threshold too high and accidentally eroding thin, legitimate details like hair or fine lines.
Expecting this to fix color fringing — it only removes isolated pixels based on alpha, not color.

Tips

Run this before the edge softener tool for the cleanest final result: despeckle first, then feather what remains.
Use the transparency checker tool afterward to confirm how much of the image's non-opaque area changed.

PNG Edges Cleaner