Compressing a Scanned Book: What Each Level Does to 60 Pages

On a scan, text quality means legibility

In April 2026, a question on a PDF forum asked how to bring a 580-page, 33 MB book scan down to about 25 MB and keep the text quality. On a scan, the words are pictures of letters, so text quality means one thing: legibility once the pictures shrink.

On October 20, 2026, in Chrome on a Mac, we ran Compress PDF on a 60-page grayscale scan at 200 DPI. Each page is a JPEG of about 1714 × 2210 pixels and about 770 kB, and the tool shows the file as 45 MB, or 47.1 million bytes.

What compression changes on a scanned page

Compress PDF works on the page pictures. Each one is resampled if it is finer than the level allows, then saved again as a JPEG at the level's quality, and the new copy is kept only if it is at least a tenth smaller. A hidden OCR text layer, links and bookmarks are copied as they were. Under More settings, Turn images to grayscale drops the color from a color scan, and metadata such as title and author is cleared unless you switch that off.

Some pictures stay as they are on every level: one-bit black-and-white pages, whether JBIG2, CCITT or plain, and JPEG 2000. All of it runs in your browser tab, and the scan stays on your device.

Three levels on the same 60 pages

Best quality sets a 300 DPI ceiling and quality 85, so our 200 DPI pages kept every pixel and were only re-encoded. That alone made the file a fifth smaller: 36 MB, a little over 600 kB a page, in about 30 seconds.

Balanced sets a 150 DPI ceiling and quality 65. Pages became 1285 × 1657, a little over half the pixels, at about 175 kB each. The file came to 10 MB, 77 percent smaller, about 31 seconds after we dropped it in. Our three-page check on October 20, 2026 showed the same drop: a 2.3 MB scan came out at 539 kB on the balanced level.

Smallest file goes to 96 DPI and quality 45: 823 × 1061, under a quarter of the pixels. The file came to 3.9 MB, 91 percent smaller and under 70 kB a page, in about 17 seconds.

What the text looks like

Under the preview, “Close-up of page 1 at 100 %” shows the same patch before and after, at its size on paper. On page 1, the text stayed readable at 150 DPI and at 96 DPI.

For Smallest file, the tool warns that small text on scans may become unreadable, so check footnotes, page numbers and tables in the close-up, not only headings.

With a file this size, once a build takes longer than three seconds, a new level no longer rebuilds by itself: the tool shows “Result is out of date” until you press “Update result”.

Target size, from 2 MB to a 25 MB cap

Target size starts from a number. The tool first measured the floor, “The smallest this file can get: 1.4 MB”, and locked lower targets, so 1 MB was not available. With 2 MB, the search took about two minutes and returned 1.8 MB. Smallest file had already given 3.9 MB, so that result sits at 72 DPI, below every fixed level: check the close-up before you send it.

The search begins at full resolution with quality 65, then 45, and steps the resolution down only if the file still does not fit. Where the file first fits, it raises the quality again until the file lands within a tenth under the target, in eight builds at most. So a mild cut, like 33 MB to 25 MB, keeps every pixel whenever quality 45 at full resolution fits. A custom target is typed in kilobytes; leave a margin, such as 24000 for a 25 MB cap.

While it searches, Target size holds every decoded page in memory; a fixed level handles one picture at a time. On a long book, a fixed level is the lighter choice. If the browser runs out of memory, the tool says so and offers Split PDF.

When OCR matters, and when to split instead

Compression never makes a scan searchable. To search or copy the words, run OCR PDF first, on the full-resolution scan: recognition reads the same pixels that compression thins out. It runs on our server: the file is uploaded, deleted after processing, and the result is wiped within about two hours. One file takes up to 500 pages and 300 MB, so a 580-page book goes in two parts: cut it with Split PDF, recognize each part, join them in Merge PDF, then compress.

Splitting suits pages that are already lean, if the destination takes several files. The forum book carries about 57 kB a page, less than ours after Smallest file; if those are one-bit pages, every level returns about the same size. Split by size cuts a PDF into parts under a limit, such as the ready 25 MB, in page order and without re-encoding: the one route here that leaves every pixel as scanned.

Compress PDF is one of more than 200 free tools on PulsarKit, with no account and no subscription. It takes files up to 300 MB; it does not recognize text and has no black-and-white mode.