Multipage TIFF, LZW Compression and Layers: How It Works

Anatomy of a TIFF: header → IFD → pixels

Every TIFF reads by one simple scheme: a tiny header points to a directory, and the directory points to the pixels. Just three levels — but it's exactly this chain of pointers that makes the format so flexible.

A file opens with an 8-byte header. The first two bytes are the byte order: II (Intel, least-significant byte first) or MM (Motorola, most-significant first). Next comes the "magic number" 42 — a signature that tells a program it's really looking at a TIFF and not something else. The last four bytes of the header are an address: the offset to the first image directory inside the file.

That directory is called an IFD (Image File Directory). Think of it as the table of contents for one image: it holds no pixels itself, but lists what the image is and where its data lives. The IFD is laid out very regularly: first a two-byte count of how many entries it has, then exactly 12 bytes per entry (these are the tags), and right at the end, 4 bytes pointing to the next IFD (or zeros if it's the last one).

Tags: why the format is "tagged"

The T in TIFF stands for Tagged, and that's the whole point. Every property of the image is recorded as a separate tag — a short "property → value" record. Width? A tag. Height? A tag. Compression type, DPI resolution, color model, orientation, bits per channel, the addresses of the pixel data — all of these are separate, numbered tags.

Each tag is exactly 12 bytes and has four parts: the tag number (which property this is), the data type (number, string, fraction), the count of values, and finally the value itself or an address. The clever bit: if the value fits in 4 bytes, it sits right inside the tag; if not, the tag holds an address pointing to where the value is stored elsewhere in the file. That keeps the directory compact and stops long data from bloating it.

This "taggedness" is what makes TIFF nearly immortal. A program reads the tags it knows and quietly skips the ones it doesn't. So a file from the 1990s still opens today, and new capabilities (a new compression type, geographic coordinates, a color profile) are added simply as new tags — old software doesn't understand them, but it doesn't break either.

Strips and tiles: how pixels are laid out

TIFF doesn't store the pixels as one solid block but split into pieces. That's handier: a program can load just the part of a huge image it needs without reading the whole file, and each piece is compressed independently. There are two ways to split.

• Strips. The image is cut into horizontal bands — a few rows of pixels each. This is the basic, most common method; any program can open a strip-based TIFF.
• Tiles. The image is cut into equal-sized rectangular tiles, like a mosaic. This is more efficient for very large images — maps, satellite imagery, scans of hundreds of megapixels: to show a fragment, you only read a few tiles.

The addresses and lengths of these pieces are tags too (one set for strips, another for tiles). From them a program knows exactly where each band or tile sits in the file and how many bytes it takes.

Multi-page is a chain of IFDs

Now the fun part — how dozens of pages fit into one .tif. The secret is already hidden in the anatomy: remember those 4 bytes at the end of every IFD pointing to the next directory? That's the multi-page mechanism.

Each page is a separate IFD with its own tags and its own pixels. The first page points to the second, the second to the third, and so on. The last IFD holds zeros instead of an address — a signal that says "nothing follows." The result is a singly linked list, like railway cars coupled one to the next.

An important detail: the pages are independent. One can be a color photo scan in LZW, the next a black-and-white page of text in compact CCITT compression, each with its own resolution and size. That's exactly why scanners and fax machines love TIFF: a whole multi-page contract leaves as one file, with each sheet packed optimally inside.

LZW compression in plain English

Inside the strips and tiles, the pixels are usually compressed. The most famous method in TIFF is LZW (Lempel–Ziv–Welch), the same algorithm once used in GIF. The headline fact: it's lossless. The decompressed file matches the original byte for byte — not a single pixel suffers, unlike JPEG.

LZW works like a smart dictionary. Imagine you're copying out a text and notice the word "image" appears a hundred times. Instead of writing it out each time, you define a shorthand once — say, "#7" — and from then on write only that. LZW does the same with sequences of bytes: it finds repeating chains, adds them to a dictionary and replaces them with short codes. The more repetition in the data, the stronger the compression.

The elegant part is that the dictionary doesn't need to be stored in the file. Both the packer and the unpacker build it on the fly, by the same rules, reading the data in order. So when you open the file, the dictionary rebuilds itself, and the picture reassembles exactly as it was.

There's a subtlety worth knowing. On flat areas (logos, diagrams, text, black-and-white scans) repetition is plentiful — LZW shines. But on noisy photographs, neighboring pixels are almost always slightly different, repetition is scarce, and the gain is modest. Worse, on 16-bit images, naive LZW can sometimes even grow the file. The rescue is a predictor — a trick where, instead of the pixel values themselves, you store the difference from the neighbor. On smooth gradients the neighbors are close, the differences come out small and uniform, and now both LZW and ZIP compress noticeably tighter.

Other compression modes and when to use each

LZW isn't the only option. TIFF was designed as a "container," and the compression type is just the value of one of its tags. Here are the main modes and what each is good for.

A few rules of thumb, no formulas. CCITT G4 is the champion for compressing black-and-white text: a multi-page scanned contract weighs almost nothing in it, which is why scanners reach for it. ZIP with a predictor is the best default for photographs and deep color. LZW is for when you need maximum compatibility with older programs. And JPEG-in-TIFF exists, but it runs against the very idea of the format: if you're going to sacrifice quality anyway, it's simpler to save a separate JPG from your TIFF.

Layers, transparency and alpha

TIFF can store more than just color channels. People often muddle this up, so let's sort it out — these are two different mechanisms.

Transparency (the alpha channel). Beyond the color channels (RGB or CMYK), TIFF adds "extra samples" — the ExtraSamples tag. One of them is alpha: for each pixel it sets the degree of transparency. It comes in two flavors. Unassociated alpha stores color and transparency separately — the familiar mask. Associated (premultiplied) alpha has the color already multiplied by transparency; this makes compositing layers faster, but it has to be accounted for during conversion, or you'll get a dark or light fringe around the edges.

Photoshop layers. Full layers (as in a PSD) are stored more cunningly. The image itself is always kept flattened — that's what every program sees. The individual layers, Photoshop tucks into a private ImageSourceData tag. Open such a file in Photoshop and the layers are there; open it in any other viewer and you'll see only the final picture.

BigTIFF and GeoTIFF: giants and special editions

Classic TIFF has a hard ceiling. All addresses inside the file are 32-bit, which means the file can't be larger than 4 gigabytes: beyond that, the addresses simply run out of bits to point at the byte you need. For a photo that's endless room, but for a satellite image of a whole country or ultra-high-resolution microscopy, it's tight.

BigTIFF lifts the limit. It's essentially the same TIFF, but with 64-bit addresses; and the magic number in the header changes from 42 to 43 — by which a program instantly knows it's looking at the "big" version. The theoretical ceiling rises to astronomical figures (millions of petabytes), so you can forget about 4 GB. From the outside it's all the same — the same tags, the same IFD logic — just wider addresses.

GeoTIFF solves a different problem — not size, but meaning. An ordinary picture doesn't know which place on Earth its pixels correspond to. GeoTIFF adds a few special tags with geographic referencing: the pixel scale on the ground (ModelPixelScale), reference points (ModelTiepoint) and a description of the coordinate system (GeoKeyDirectory). Thanks to the format's "tagged" nature these are just new tags — old programs ignore them and still show the picture, while GIS software reads them and pins the image precisely onto a map. GeoTIFF is the de facto standard for satellite imagery, elevation maps and aerial photography, and it's often paired with BigTIFF, because such data easily steps past 4 GB.

What this means in practice

Knowing the internals isn't theory for its own sake. Here's how it helps with everyday tasks.

• You understand why TIFF is heavy. Lossless compression doesn't discard data, it only hunts for repeats. So a photo in TIFF is several times bigger than a JPG — and that's normal for a master file.
• You choose compression deliberately. Text and scans — CCITT G4; photographs — ZIP with a predictor; need maximum compatibility — LZW. It directly affects file size.
• You don't lose pages in conversion. Knowing that multi-page is a chain of IFDs, you won't be surprised that a good converter keeps every sheet. For a document, pick TIFF to PDF; for a single picture, TIFF to JPG.
• You're careful with layers and alpha. Layers live only in Photoshop, and premultiplied alpha has to be handled correctly or you'll get a fringe around the edges.

Don't need to assemble a document, just a lightweight copy of a single picture for the web or sharing? Then TIFF to JPG is the way. And the full list of conversions is always on the all formats page.