Archives
Digitising a newspaper archive: the decisions that cannot be undone
A digitisation project is mostly logistics, and about six decisions. The logistics can be corrected as you go. The decisions cannot, because reversing one means handling every original again — and handling is the part that costs money…
This is a guide to the six, and to which way we would err on each.
Capture once, at the highest quality you can afford
The governing principle: the master is captured once and everything else is derived from it.
That makes the master's specification the most consequential decision in the project. Too low and the archive is permanently limited by a choice made for a storage budget that stopped mattering years ago. Too high and the project costs more and delivers more slowly.
Err high. Storage gets cheaper every year; re-scanning never does.
Resolution
Resolution has to be stated relative to the original, not as a number in isolation. The same setting produces very different results on a broadsheet page and a 35mm negative.
The working test is whether the smallest meaningful detail is captured with enough pixels to be read. For newsprint that means the body text — including the small type in classified columns and picture credits, which is where the detail actually lives. For negatives it means grain structure.
Scan a representative sample at two or three settings and read the results at full size before committing. This takes an afternoon and prevents the most expensive class of mistake.
Bit depth and colour
Three choices, in increasing cost: bitonal, greyscale, colour.
Bitonal — pure black and white — produces small files and is adequate only for clean text on clean paper. On newsprint it destroys halftone images entirely, and the photographs are usually the reason the archive is worth digitising.
Greyscale preserves halftones and tonal variation. For a monochrome archive it is generally the right answer.
Colour captures paper tone, ageing, ink variation and anything printed in colour. It costs more in storage and in processing, and it records the physical state of the object, which a greyscale scan does not.
Where a page has any colour at all, or where the physical condition is part of the record, scan in colour. Reducing colour to greyscale later is trivial; the reverse is impossible.
File format for the master
The master should be uncompressed or losslessly compressed, in a format that is openly documented and widely readable.
Avoid any lossy format for the master, however good it looks. The loss is invisible at first and compounds through every subsequent processing step, and it cannot be undone.
Avoid proprietary formats tied to one vendor's software. The archive will outlive the software; the decision that matters is whether the file can be opened in thirty years by somebody with no access to today's tools.
Derivatives — the compressed versions people actually view — are generated from the master and can be regenerated in any format at any time.
Metadata, created during capture
Metadata created at capture is cheap. Metadata created afterwards is a second project, usually unfunded, and it is the reason many digitisation efforts produce a pile of images nobody can search.
The minimum per scanned item: what it is, its date, its source publication, its page or item reference, the physical object it came from, the date of scanning, and the equipment and settings used. That last group is technical metadata and it matters more than it appears — it is what allows a later reader to judge whether an image is a faithful capture or an artefact of the scanner.
Descriptive metadata beyond that — subjects, people, places — is where the real work is. Budget it as part of the project, not as a phase to be scheduled later, because later never arrives.
Naming and identifiers
Each item needs an identifier that will never change and never be reused, and it should be assigned at capture rather than derived from something mutable.
An identifier built out of a shelf location breaks when the material is rehoused. One built out of a publication title breaks when a title is renamed retrospectively. A plain sequential identifier with the context held in metadata survives both.
Where projects actually fail
Not on the scanning. On three things: metadata deferred until after capture and then never done; a master specification chosen for the storage budget of the project's first month; and no plan for what happens to the physical originals afterwards.
The originals question deserves an early answer. Digitisation is not a substitute for keeping them — a scan is a record of an object, not the object — and a project that assumes disposal will be argued about at the worst possible moment.
Conclusion
Capture once, err high on resolution, use colour where there is any colour or any physical condition worth recording, keep the master lossless and openly documented, create technical and descriptive metadata during capture rather than after, and use identifiers that do not depend on anything that can change. Every one of these is cheap to decide now and requires re-handling every original to fix later.