File formats for archival photography
A photograph of an object is rarely one file. There is what the camera wrote, the picture developed from it and kept for the long term, and the copies made for a web page, a report or a loan request. Each has a job, and each job has a format that suits it. Confusing them is costly: a collection that keeps only its copies has lost quality it cannot get back.
Original, master and copy
- The camera original is the file as the camera wrote it: a raw file, untouched. It holds everything the sensor recorded.
- The preservation master is the picture developed from it and kept for the long term: in a defined colour space, uncompressed or losslessly compressed. It is what every copy is made from. Once written, it is not rewritten.
- The access copy is made from the master for a particular use: smaller, compressed, in the colour space of screens. It can be made again at any time, so it carries nothing that cannot be lost.
Metamorfoze states the idea of a master precisely: an unprocessed, first-generation file with its colour space embedded. Stitching and focus stacking are kept out of it and left to the copies.
RAW
A raw file holds the sensor's data before it has been turned into a picture: before demosaicing, white balance and tone curve. It keeps every bit the camera recorded — 12 or 14 in most cameras, 16 in some — and every decision about the picture can be made again from it.
Most raw formats are the camera maker's own — CR3, NEF, ARW, RAF — and are read by software that has been taught each one. That suits an original, and makes a poor long-term master on its own: a format nobody updates a reader for is a risk.
DNG
DNG is a raw format whose specification Adobe publishes openly. Some cameras write it directly; Apple ProRAW, from an iPhone, is a DNG. Converting proprietary raws to DNG is a common archival step, because the format is documented and widely read. Whether the maker's original raw is kept beside the DNG is a decision for each collection; keeping it costs storage and loses nothing.
TIFF
TIFF is the format for the preservation master. FADGI accepts TIFF or JPEG 2000 for masters, and TIFF is read by practically every image program. It can be stored uncompressed or with lossless compression (LZW or ZIP), which makes the file smaller without changing a single pixel. It carries an embedded colour profile and metadata.
In practice a TIFF holds 8 or 16 bits per channel. A 12- or 14-bit capture is written into a 16-bit file, which holds all of it without loss, and the file's metadata should say how many bits the camera actually recorded. Why, and what the standards ask, is under Bit depth.
JPEG
JPEG is the format for access. It holds 8 bits per channel and uses lossy compression: it discards detail the eye is unlikely to miss, and discards more each time it is edited and saved again. In sRGB, it shows the same on almost every screen and in every browser.
That makes it right for publishing and wrong for keeping. FADGI gives a JPEG no level as a master at all. Metamorfoze accepts JPEG, but in sRGB it reaches Extra Light at most, the level for text documents kept as carriers of text.
HEIF and JPEG 2000
HEIF, the format many phones save in, can hold 10 bits and compresses better than JPEG, but it is less widely read and is lossy in ordinary use. JPEG 2000 can be lossless and is accepted by FADGI for masters; it is used by some large digitisation programmes, but fewer everyday programs open it.
What to archive, and what to publish
| Format | Bits per channel | Colour space | Changed later? | |
|---|---|---|---|---|
| Camera original | Raw or DNG | As recorded | — (raw data) | Never |
| Preservation master | TIFF, lossless | As photographed: 8, or 16 for anything deeper | Adobe RGB or wider, embedded | Never |
| Access copy | JPEG | 8 | sRGB, embedded | Made again from the master |
Keep a checksum for every master and original — a SHA-256 fingerprint shows years later whether a file is still the one that was stored — and keep the record with the picture, in the file's own metadata, so that it survives the journey out of the database.
Checklist
- Photograph raw, and keep the raw as the camera wrote it.
- Develop one master per picture: TIFF, lossless, colour profile embedded.
- Never rewrite the master; make every copy from it.
- Publish JPEGs in sRGB, and make them again rather than editing them.
- Record a checksum for every file that is kept.
The PHOTARCH app exports the master as a TIFF in Adobe RGB at the depth the picture was photographed in, with a JPEG beside it as the access copy, and never rewrites the original it imported. In PHOTARCH's own levels a JPEG master is a record shot at most: a study asks for a raw with at least 10 bits recorded and a lossless master, a reproduction for 12 bits and a 16-bit TIFF in Adobe RGB or wider.