TIFF: What Archives Scan To, and What Comes Out the Other Side

A TIFF is what you get when a scanner is told not to lose anything. Museums, land registries, hospitals and legal archives all standardised on it, and the reason is structural rather than sentimental: the format describes whatever the pixels happen to be instead of assuming a particular kind of picture.

That flexibility is also the problem. Two files with the same extension can differ from each other more than a JPEG differs from a PNG, and a reader that opens one may refuse the other outright. Working out which kind of TIFF you actually have is most of the job.

Tagged image file directories#

The name is literal. A TIFF opens with a two-byte order mark, II for little-endian or MM for big-endian, and then a pointer to an image file directory. That directory is a list of numbered tags: width, height, bits per sample, compression, photometric interpretation, and where the pixel strips sit in the file. A reader walks the tags it understands and is meant to skip the rest.

Because one directory can point at the next, a single file can hold many images. That is what people mean by a multi-page TIFF, and it is how a forty-page scanned contract arrives as one attachment. The .tif and .tiff spellings are the same format; the three-letter form is a leftover from MS-DOS filename limits, and renaming between them changes nothing at all.

Multi-page files and the page you actually get#

Converting a multi-page TIFF meets an obvious wall: PNG and JPEG hold exactly one image each. Convert TIFF to PNG and convert TIFF to JPG both export the first image in the file. The other pages are not damaged and not altered. They are simply not in the output.

The compression variants#

TIFF does not have a compression scheme. It has a tag that names one, and the list is long. Every option below is legal, and every one of them produces a file called .tiff:

• None. Raw pixel strips: large, predictable, and the safest thing to hand a simple reader.
• PackBits. A run-length scheme from the early Macintosh era. Cheap, lossless, and close to useless on photographs.
• LZW. The same dictionary method GIF uses. Effective on flat artwork and common in prepress work.
• Deflate, often labelled ZIP. The method PNG uses, and usually the strongest of the lossless options.
• CCITT Group 3 and Group 4. One bit per pixel, built for fax and document scanning. Very small files, bilevel only.
• JPEG inside the TIFF wrapper. Lossy pixels in an archival container, which rather defeats the point of choosing TIFF.

The decoder here reads top-left oriented 8-bit RGB and RGBA images with contiguous strips, stored uncompressed or with LZW, Deflate or PackBits. Anything outside that set is refused with a message rather than converted badly: bilevel fax scans, CMYK prepress masters, 16-bit and floating-point scientific data, tiled layouts, planar channel storage and BigTIFF all stop at the door. That is deliberate, and it follows the principle in verifying conversion results. A refusal you can read beats a silently wrong colour conversion you cannot.

What does not survive#

• Extra pages. One image comes out, whatever went in.
• Embedded ICC profiles. Output colour is written as ordinary sRGB, so a contractual prepress match is not guaranteed.
• EXIF, XMP and IPTC blocks, including scanning provenance, operator notes and rights fields.
• Nothing beyond that is silently degraded. A 16-bit archival scan is not quietly reduced to 8 bits behind your back, because a file at that depth is refused rather than converted.
• CMYK separations stop at the door for the same reason, rather than being approximated in RGB and handed back looking plausible.

None of that is an argument against converting. It is an argument for keeping the TIFF and treating the result as a delivery copy, which is exactly the habit described in PSD, EXR and HDR. The TIFF format reference sets out the variants in more detail.

Picking a target#

For a scanned photograph going to email or a website, JPG at a high quality setting is usually right and is dramatically smaller. For text, line art, plans or anything with large flat areas, PNG is the better answer, because JPEG artefacts cluster exactly where the edges are. Image formats compared sets the trade-off out in full.

Going the other way is possible too, and worth one caveat. Producing a TIFF from a web image gives you the container and the lossless storage, but it cannot invent the bit depth, the colour profile or the scan resolution that the original never carried. A TIFF made from a JPEG is a lossless copy of a lossy image, not an archival master.