How to Remove the Background From a PDF Page

Removing the background from a PDF page fails in one specific way, and almost every tool handles it the same wrong way. You drop in a document, the tool reports that it converted the file, and you get a JPG with a white box where the background used to be — or a sign-up wall, or a paywall on the download, or a page that says your file is being "securely handled" by a server you cannot see. You conclude the tool is broken, or that you did something wrong, and you try a different site that does the same thing.

That is the failure worth understanding, so this page is mostly about it: how to strip the page background without uploading anything, what the output really is, and the numbers from a real run. The file never leaves your machine.

What a PDF page actually is

A PDF does not store a page as a picture with a background layer underneath. It stores drawing instructions: put this glyph at this coordinate, draw this rectangle at this size, place this image at this position. A background is something a reader perceives from the arrangement — the file has no object that says "this is the background", and no object that says "everything else is the subject".

Removing the background therefore means rendering the page to pixels, deciding which of those pixels are the background, and making them transparent. That is a different job from extracting text or converting to JPG, and it is why a PDF-to-JPG converter will always hand back a white rectangle: JPG has no alpha channel, so there is nowhere for transparency to live.

A scan makes this harder, not easier. A scanned page is one large image — a photograph of paper — and the "background" is the colour of that paper, which may be cream, grey or yellowed rather than stark white. The tool has to learn the colour from the page itself rather than assuming white.

Measured: what actually comes out of a real document

I ran a real 3-page A4 document through the tool in Chrome. Each page was rendered at twice its size in the PDF — 144 DPI — and the background was stripped from the pixels. The row reported 3 pages · 87% of pixels were background · 1.2 MB → 84 KB, and the whole run took 2500 ms from the file landing in the page to the result line appearing. No long task over 50 ms was recorded while it ran, so the page never froze.

The downloaded ZIP contained three transparent PNGs — report-page-1.png, report-page-2.png, report-page-3.png — each 1190 x 1684 pixels with an alpha channel. A pixel sampled from the corner of page 1 read [0, 0, 0, 0]: fully transparent. A pixel sampled from the middle of the page, over printed text, read [43, 43, 43, 255]: fully opaque, just as it should be.

A single-page export behaves the same way but without the ZIP: one page downloads as a single {name}-page-1.png, no archive wrapper.

What the tolerance actually does

The engine samples the four corners of each page to learn what colour the background is, then walks every pixel and zeros the alpha of anything within tolerance of that colour. The default is 30, which catches a stark white page and the off-white edges around it.

A scanned page is the case the tolerance exists for. I ran a single-page scan whose paper was cream rather than white through the tool at the default tolerance, and the corners stayed opaque — the engine had learned "cream" as the background, but the cream was not close enough to white at tolerance 30 to be stripped. Turning the tolerance up to 60 made the corners transparent and left the text intact. That is the whole control: the page teaches the tool its own background colour, and the slider decides how forgiving the match is.

What the output really is

Transparent PNG. That is the only format a browser canvas can write that keeps an alpha channel, so the output is always PNG and the background stays transparent. JPG and WebP are not offered because both would silently flatten the alpha to white, and the visitor would get a white box where the background used to be — exactly the thing they came here to avoid.

Every page is rasterised, so text in the PNGs is no longer selectable or searchable. Remove the background only if you need pictures, not if you need words. The PDF you dropped is untouched, so you can come back with a different tolerance at any time.

What the competition does

Two claims, both fetched from the pages themselves on 2026-11-03 rather than remembered.

Adobe's converter, under its drop zone: "Your file will be securely handled by Adobe servers and deleted unless you sign in to save it." That is a document going to a third party, and a sign-in between you and the download. For a contract, a payslip or a medical letter, the upload is the part that matters.

One of the "free" converters says this about the rest of the field: "most free tools fail completely on scanned documents, leaving you with empty output." That is the industry default, stated by a competitor, and it is the failure this tool is built around. The tolerance slider exists precisely because a scan does not have a stark white background, and a tool that assumes white leaves the corners opaque.

The other half of the category is the same tool with a task limit: two documents a day, or a queue, or a paywall on the batch. None of that applies here, because nothing is uploaded and there is no server to meter. You can disconnect from the internet after the page has loaded and the background removal still works.

The short version

A PDF page has no background layer to strip — it has drawing instructions, and the background is what the reader perceives. This tool renders each page to pixels, learns the background colour from the corners, and makes everything within tolerance transparent. The output is transparent PNG, one file per page, and nothing is uploaded anywhere.

And none of it requires sending your document to a server.