Is Pixelation Reversible? Use a Solid Black Box, Not Blur
When you share a screenshot or photo with a name, number or address in it, pixelating or blurring it feels like enough. Public research and tools show that, depending on how it was done, the original text can sometimes be guessed. This page separates what we could confirm from what we could not, and explains how to hide information with an opaque fill instead.
What can happen with pixelation
Sources we checked (checked 2026-11-09).
- Depix (GitHub, spipm/Depix): described as a proof of concept for recovering plaintext from pixelated screenshots. It was archived in April 2026 with a pointer to a successor. We did not read the README's technical conditions, so we do not describe them here.
- Unredacter (Bishop Fox, bishopfox/unredacter): the README presents it as a tool showing that pixelation should not be used instead of a black box. To use it you cut out the pixelated part, find the block size, match the style (spacing, weight) to the original text, and supply candidate characters. The README says this takes a lot of manual work.
So not just anyone can automatically read pixelated text. But the README's procedure leaves room for educated guesses, which is why pixelation is a poor fit for information you want hidden. We did not check under which conditions, or how much, can be read.
What can happen with blur
Deconvolution (Wikipedia: Deconvolution, checked 2026-11-09; an encyclopedia entry, not a primary source) is described as a way to sharpen images blurred by motion or camera shake. The same article says estimates get worse when noise is high (low signal-to-noise ratio).
How much can be recovered varies a lot with the blur type, strength and image compression. We found no source showing how well it works on personal-information text.
Why an opaque fill is safer
- With an opaque solid color, the original pixel data in that area is gone from the image. No clues (block averages, outlines) remain to guess from.
- This assumes you export a new image. If you only put a shape over the picture in an editor and send a file that keeps the original layer, nothing is hidden.
- Cover a little more than the text so no edge of a character is left.
Tools
Our image fill tool (/image-redact/) has a Japanese-only interface and there is no English version yet. It draws the image on a canvas, fills the chosen areas with opaque black or white, and re-exports the canvas as a PNG or JPEG, with no mosaic or blur option. It runs in your browser and nothing is uploaded. If you need an English interface, there is a workaround: turn the image into a PDF with the image-to-PDF page (it accepts PNG and JPG), then black out areas with the PDF redaction page, which rebuilds redacted pages as images. The result is a PDF, not an image file.
Steps
- Load the image into a tool that fills areas with an opaque color and re-exports the image.
- Drag over the part you want to hide, a little larger than the text.
- Open the saved file and check with your own eyes that nothing is readable and no edge is left.
Black boxes on PDFs have their own pitfall: the text can stay in the file. See the PDF redaction page.
FAQ
Is pixelation reversible? Can pixelated text be recovered?
Under the right conditions it can be guessed (Depix, Unredacter). It is not guaranteed, but it is not suited to information you need hidden.
Can blurred text be undone?
Deconvolution can sharpen some blurred images (Wikipedia), but accuracy drops with noise. We could not confirm whether it works on personal-information text.
How do I hide information safely?
Cover it with a fully opaque solid color and export a new image.
Is a black box on a PDF the same?
Not quite: a rectangle on top can leave the text under it in the file. See the PDF redaction page.
Related: How to redact a screenshot so it cannot be seen through / Redact a PDF (no upload) / Photos to PDF