Pixelation vs Blurring vs Black Bars

Every redaction method throws away information. The question that decides whether it worked is how much it threw away, and whether what remains is enough for someone to reconstruct or guess the rest.

The short answer

• Solid masking (a black bar or filled shape) is the only one of the three that is unconditionally safe. The pixels underneath are replaced with a flat color. Nothing about the original survives.
• Pixelation is strong for faces and general scenery, risky for text and short codes. It destroys detail, but the remaining blocks still carry average color information.
• Blurring is the weakest of the three, and its weakness is not obvious from looking at it.

Why blurring is weaker than it looks

A Gaussian blur spreads each pixel's value across its neighbors. Crucially it does not discard those values, it redistributes them. The information is smeared rather than deleted, and mathematically a blur is a convolution that can be partially inverted if you can estimate the blur radius.

A mild blur at a small radius is genuinely reversible to a useful degree. A heavy blur at a large radius, applied to an image that is then re-encoded and downscaled, is much harder to undo, but "harder" is doing a lot of work in that sentence.

Blurring also fails badly against machine learning approaches. Models trained on paired blurred and sharp images can produce a plausible reconstruction. It is important to be precise about what that means: the model is generating a face consistent with the blur, not recovering the true one. That is little comfort in practice, since a confident and roughly correct guess is often all an attacker needs.

Where pixelation sits

Pixelation is genuinely destructive in a way that blurring is not. Each block is collapsed to a single averaged color, and every variation within that block is gone from the file permanently. A 32 by 32 pixel region becomes one value. That is a real, irreversible loss of over a thousand measurements.

Its weakness is a different shape. Pixelation is deterministic: the same input and block size always produce the same output. So if you can guess what was underneath, you can check your guess. Generate a candidate, pixelate it identically, and see whether the blocks match.

This is why the answer depends entirely on what is underneath. A human face is effectively unbounded in variation and cannot be brute-forced this way. A six-digit account number in a standard font has a million possibilities, which a computer will exhaust in seconds. Bishop Fox released a tool called Unredacter in 2022 that does precisely this to pixelated screenshots, and we cover how it works in detail here.

Choosing between them

Use solid masking when

• You are hiding text, numbers, codes, email addresses, license plates, or anything else from a finite set.
• The material is legally sensitive or you cannot afford to be wrong.
• You do not need the viewer to understand what was there.

Use pixelation when

• You are hiding a face, a body, or a scene, where the range of possibilities is effectively unlimited.
• You want the viewer to see that something was deliberately obscured, which pixelation signals more clearly than a blur.
• You want to preserve the composition and rough context of the photo rather than punching a hole in it.

Use blurring when

• The goal is aesthetic rather than protective, such as de-emphasizing a background.
• The hidden content is not actually sensitive and you only want it to be less distracting.

Combining methods

For a document containing both a photo and written details, the right approach is usually mixed: pixelate the face so the image still reads naturally, and place solid bars over the name, number, and address. Using one method uniformly across a page is what leads to the weakest element determining the outcome.

Whichever you choose, export to a flat format such as PNG or JPG and share that file. Layered formats and some PDF redactions keep the original content in the file underneath the covering element, which has caused a long series of public disclosure failures.

Our pixelation tool handles the pixelation side in your browser, and our guide on reversing pixelation goes deeper on where it breaks down.