How We Restore a Scan: Before and After

A scan is not a print file

A library photographs a sheet to record it, not to print it. What comes out is honest and usually unusable: the scanner’s own colour profile baked into the file, a dark mount around the paper, a colour bar in one corner, a grey field that slopes from one side to the other because the lights were not even, and a hundred and fifty years of dust sitting on top of the engraving.

Restoration here means taking that file and making it printable without making it a different sheet. The second half of that sentence is the hard part, and it is the part this page is about.

What the engine does

Every map goes through the same six operations, in the same order, with the same limits. Nothing is done by eye and nothing is done twice.

• Colour profile to sRGB. Some archive scans carry the scanner’s own profile — one holder ships files tagged Metis DRS 2A0 CC24. Without the conversion every colour on your screen is wrong.
• Crop of what is not the sheet: the dark mount, the colour bar, the table underneath. The crop is refused outright if it would remove more than a quarter of the image — that rule exists because the detector once mistook one half of a two-sheet map for the whole map.
• Black point and white point. The black point is read at the 0.5th percentile, the white at the 99.5th, and the result is placed inside 2…248 so that nothing is clipped at either end. The black point is capped at 25 no matter what the percentile says: on a pale engraving, subtracting more than that starts printing the back of the sheet.
• Stretch capped at ×1.25. A sheet with a genuinely flat tone — a range under 30 — is left alone rather than pulled apart.
• Evening of the paper field, so the light gradient left by the copy stand stops reading as a stain.
• Dust and specks removed one by one, counted, and the count written down. On the last celestial sheet that number was 66 specks; on sheets with foxing it runs into the hundreds.

Saturation is nudged by six percent only when the paper is neutral. When the paper is warm — median red minus blue above 12 — it is not touched at all, because warm paper plus saturation is how a sheet turns amber.

The six things we never do

These lines are printed on the passport that ships inside every archive you download, for every map, without exception:

The third one is worth pausing on. A map that has been stretched to fill a frame is no longer a record of anything. Our formats add paper at the edges — extended with texture taken from the same sheet — rather than distorting the map to reach 3:2 or A2.

The second one costs us sheets. Where a previous owner has written over the engraving, or where a stamp sits on the printed image rather than on the margin, we do not paint it out: we either keep it, or we go and find another copy of the same map. Two sheets are in that queue right now.

How we prove we did not overdo it

Every finished map is measured against fifteen criteria, and the numbers — not the verdict, the numbers — go into the passport PDF inside your download. Three of them compare the finished file directly against the archive original, which is the only way to answer “has this been prettified?” honestly.

Here is what those three actually say across the 53 sheets restored so far:

The ink figure is the one to read twice. A median of 1.00 means that on a typical sheet the amount of ink in the finished file is the same as in the archive scan — the line was cleaned around, not drawn over.

Thirteen out of fifteen, not fifteen

The threshold to publish is 13 of 15, not a clean sweep. Twenty-eight of the fifty-three sheets pass all fifteen; the rest carry a named miss, and the passport says which one.

We prefer that to the alternative. A sheet can fail “field evenness” because the original paper genuinely is darker on one side, and forcing that number to pass would mean flattening a real property of the object. When a criterion is missed for a reason that belongs to the sheet rather than to our work, it is recorded and the map is published. When it is missed because of our work, the map goes back.

Six further criteria are checked by eye rather than by arithmetic: that the printed image is not cut off, that a join between two scanned halves is not torn, that pencil marks on the margin are accounted for, that no correction reads as a rectangle, that the trim frame is straight, and that specks and stains are counted individually. These are the ones a number cannot settle.

What you can see for yourself

Every listing carries a before and after image: the archive scan and the finished file side by side, the same detail at 1:1, no scaling trick between them. It sits at the end of the gallery rather than the start, because the first thing worth showing you is the sheet itself.

Between those two you will find seven crops at full resolution — the title cartouche, the imprint line, the decoration, the densest hatching on the sheet. Those are not illustrations of our work; they are the file you will print, shown at the size you will print it. If the line holds up there, it holds up on the wall.

What restoration cannot do

It cannot invent resolution. Every listing states two numbers — the map and the full sheet with margins — and both come from the archive scan measured at 300 dpi. If a sheet only reaches 38 cm, we say 38 cm.

It cannot make a damaged sheet undamaged. A tear that crosses the engraving stays; a missing corner stays missing. What we remove is dirt, not history.

And it cannot turn a bad scan into a good one. Roughly a hundred sheets in our own queue are flagged for exactly this reason — the copy we have is too small — and the answer to those is to find a better copy, not a better filter.