What Is DjVu? The Format That Compresses Scanned Books Better Than PDF

What DjVu is in plain English

DjVu (pronounced "déjà vu," the French phrase for "already seen") is a format for storing scanned documents: books, magazines, old textbooks, drawings. Its single goal is to make a scan of a paper page take up as little space as possible while staying readable.

If PDF is a universal "container for everything" (text, layout, forms, signatures), DjVu is a narrow specialist. It can barely do anything except one thing: show scans. But it does that so well that a book page in DjVu often weighs 5–10 times less than the same scan saved as an ordinary PDF or a JPEG.

Where it came from: scientists at AT&T Labs

DjVu didn't come from a corporation chasing profit — it came from a research lab. It was developed at AT&T Labs in Red Bank, New Jersey, between 1996 and 2001. Among the authors are names the whole AI world now knows: Yann LeCun and Yoshua Bengio (both later Turing Award winners), along with Léon Bottou, Patrick Haffner and others.

The problem was very specific for its time. It's the late 1990s, almost nobody has broadband, and pages crawl in over a modem. Yet people want to digitize entire libraries. A good-quality book-page scan (300–600 DPI) in JPEG weighed hundreds of kilobytes — putting that online was painful. The world needed a format that would shrink scans dozens of times over without turning the text to mush.

The core idea: three layers per page

Here is the genius of DjVu. The authors noticed something simple: a book page holds two completely different "things." There's crisp black text — sharp edges, just two colors. And there's the background — yellowish paper, illustrations, stains, smooth color gradients. Compressing them with one algorithm is a mistake: what works for text is bad for a photo, and vice versa.

So when encoding, DjVu takes each page apart into three layers:

• Mask — a high-resolution black-and-white layer (say, 300 DPI). This is where the text lives: the sharp outlines of letters that must stay perfectly crisp.
• Foreground — the color of the letters themselves, kept at low resolution. The ink colour barely changes from letter to letter, so it doesn't need fine detail.
• Background — the color of the paper and the illustrations, also at low resolution (around 100 DPI). Your eye won't notice fine background detail behind the text anyway.

Each layer is then compressed by its own algorithm, perfectly matched to its nature. Sharp text by one, blurry background by another. That separation is the source of the small-size magic.

JB2 and IW44: how the tiny size happens

Two different codecs do the compressing — one per type of content.

JB2 compresses the black-and-white text mask (it's a relative of the JBIG2 standard). Its trick is clever: the algorithm finds repeating shapes on the page — for instance, every letter "a" in one font and size looks almost identical. Instead of storing a thousand separate "a"s, JB2 saves the shape of the letter once, then simply records "an 'a' goes here, and here, and here." Text is endless repetition of the same glyphs, so the savings are enormous.

IW44 is a wavelet codec that compresses the color background and foreground. Wavelet compression excels at exactly this kind of smooth imagery: it drops details the eye can't tell apart and barely touches quality where it matters. Since the background is stored at low resolution, it compresses down to almost nothing.

The OCR layer: a picture you can search

At first glance DjVu is just a set of page images. But the format has a pleasant superpower: a hidden text layer. When a book is digitized, optical character recognition (OCR) software "reads" the scans and attaches invisible text precisely aligned to the words in the image.

The result: you see a photo of the page with all its blemishes and period typeface, yet you can select words, copy them and search the entire book like a normal document. A picture for the eye, text for search. One caveat: not every file has this layer. If a book was scanned in a hurry without OCR, you can't search it — only flip through.

DjVu today: alive or dead?

The honest answer is somewhere in between. Development has essentially stopped: no new versions ship, and LizardTech, the company that owned the commercial edition, wound it down long ago. A turning point came in 2002, when the Internet Archive chose DjVu for its Million Book Project and digitized millions of editions in it. But in 2016 that same Internet Archive announced it would no longer create new DjVu publications: browsers had dropped the old plugins, and the Java viewer became hard to maintain.

And yet the format hasn't vanished. Millions of already-digitized books live on in DjVu — in libraries, scientific archives, private collections. So while hunting for a rare textbook or an old monograph, you may well run into .djvu. The format isn't evolving, but it isn't dying either — it has simply become the keeper of a vast legacy.

How to open and what to do with DjVu

Double-clicking a .djvu on a fresh system usually does nothing — you need a reader app. There's no shortage: on Windows it's WinDjView or SumatraPDF, on Mac MacDjView, on a phone FullReader or PocketBook.

But if you'd rather not install a separate app for one file, there's an easier path — convert DjVu to PDF. PDF opens literally everywhere: any computer, phone, browser, e-reader. You turn a niche format into a universal one and stop depending on a special reader.

Want to dig deeper into the differences? Read our detailed DjVu vs PDF comparison or the practical guide on opening DjVu on any device. And the full list of supported conversions is always on the all formats page.