I Compared Four Archive Scans of A Apple Pie for Resolution Accuracy

One of the four public archive scans of Kate Greenaway's A Apple Pie holds a true 600 DPI at the letterform edge. The other three do not. We ran the test because we were drawing a nursery-scale alphabet reference from the 1886 edition and needed to know which source file would survive a 24-by-36-inch print without the serifs turning to porridge. What we found was not what the metadata promised. Scan resolution, as reported by the hosting institution, correlates poorly with the resolution a designer actually gets when they zoom in on a lowercase g. This piece is the audit.

Methodology: What We Measured, and What We Threw Out

We pulled the highest-resolution public download available from each of four archives that host A Apple Pie: the Internet Archive, HathiTrust, the British Library digitised collection, and Google Books. For each file we discarded the archive's stated DPI and re-derived effective resolution from the raster itself. The test used three measurements per plate: horizontal pixel span of a known letter form on the title plate, the same span rendered at a reference of the letter's physical width in the printed 1886 edition, and an edge-sharpness reading taken at the boundary of a black serif against off-white paper.

Edge sharpness is the number that matters. A file can carry 3,000 pixels across a page and still resolve a letter edge over eight or ten pixels of blur, which means the effective resolution is a fraction of the nominal number. We measured the black-to-paper transition in pixels and treated any edge that took more than three pixels to complete as compromised.

We threw out any scan with visible unsharp-mask artefacts around letterforms. Post-processing that halos the type inflates apparent sharpness and destroys the drawing. Two of the four had it. One badly.

Finding #1: The Internet Archive Scan Wins on Pixel Density, Loses on Colour

The Internet Archive file is the largest raster of the four, and its per-page pixel count is the highest we measured. On the title plate, a lowercase letter measured roughly 240 pixels wide against a printed width close to seven millimetres. That works out to something in the neighbourhood of 870 pixels per inch at the letter, which is well above the 600 DPI needed for a large-format nursery print.

The edge transition on serif corners resolved cleanly in two to three pixels, which is what we want. No unsharp-mask halo. The scan was executed on a flatbed at high native resolution and left alone, and it shows.

The problem is colour. The Internet Archive file leans warm-yellow across the whole plate, with a paper base that reads closer to a tea-stain than to the ivory of a well-preserved 1886 copy. The chromolithograph plates, which are the reason to reproduce A Apple Pie at all, come across muted and heavy. A designer can correct this in software, but colour-correcting a chromolithograph without a reference copy in hand is guesswork, and guessing at Greenaway's palette is the fast route to a wall-print that looks like a bad Instagram filter.

For a black-line reference, this is the file to use. For colour reproduction, it is the file to be careful with.

A Apple Pie The print from this article · from €29.95 View the print →

Finding #2: The HathiTrust Scan Is the Only One With a Reliable Baseline

HathiTrust provides a lower nominal file size than the Internet Archive but higher measured integrity. The letter width on the title plate resolved at roughly 190 pixels, which puts effective resolution close to 690 DPI at the letter. Below the Internet Archive number, above the print threshold, and — critical for our use — matched by a clean edge transition of two pixels.

The reason this file is our working baseline is not its raw resolution. It is the paper. HathiTrust's institutional workflow calibrates against a neutral grey card on-plate, and the paper base on this scan reads as the ivory it should be. The chromolithographs sit at plausible densities. When we sampled the red pigment on the A plate, it fell within a range consistent with an aniline red pigment behind a hundred and forty years of oxidation, not the flat crimson the Google Books file produced.

The file is uneventful. That is a compliment. A designer can drop it into a working document and trust the colour balance enough to make artistic decisions instead of correction decisions. For a chart that will end up on the wall above a cot, that trust is what separates a reference the studio uses from one it fights.

Finding #3: The British Library Scan Is Softer Than Its Metadata Claims

The British Library file arrives labelled at a resolution that would put it slightly ahead of HathiTrust. It is not. When we measured the same letter on the same title plate, the pixel width was in range, but the edge transition took five to six pixels to complete instead of two to three. The serif corners are readable, but they are rounded. Under magnification the letter looks as if it has been drawn with a slightly wet nib.

We suspect a resampling step in the archive's export pipeline — a bicubic upsize from a smaller native scan, or a JPEG re-encode at a moderate quality setting. Either operation inflates the file's stated resolution while destroying the acuity that resolution is supposed to represent. The result is a scan that will print cleanly at 12 by 18 inches and start to visibly soften at 24 by 36.

Colour on this file is the best of the four. The paper base is accurate, the chromolithograph pigments read as they should, and there is no visible cast. For a designer who needs to sample palette values or judge print density, the British Library file is the reference. For a designer who needs to trace or vectorise a letterform, it is the wrong tool. We use it for colour decisions and nothing else.

Finding #4: The Google Books Scan Is Unusable Above 8x10 Inches

The Google Books file is the smallest raster of the four. Effective resolution at the letter measured out to somewhere between 280 and 320 DPI. That is below the threshold for any nursery-scale wall reproduction. Print this file at 24 by 36 inches and the letters break into visible pixel steps at reading distance.

Edge sharpness was compromised further by aggressive JPEG compression. We measured blocking artefacts at 8-by-8-pixel boundaries in the paper areas, which is the fingerprint of a low-quality JPEG saved and re-saved through the export pipeline. Around the black letterforms these artefacts produce a faint grey halo that is not present in the original book.

Colour is the worst of the four. The paper base reads pink-white, the reds are flat and saturated in a way that suggests channel-mixed correction rather than a colour-calibrated scan, and the black plate has been crushed. Detail in the darker chromolithograph passages, which is where Greenaway's linework carries the most decorative weight, is lost.

The Google Books file exists as a searchable-text convenience, not as a designer's reference. That is fine. It should be labelled as such. A designer who does not test their source file will build a chart on top of it and discover the problem only at the print stage, which is expensive.

The Scans, Side by Side: A Comparison Table

Source	Effective DPI at letter	Edge transition	Colour fidelity	Usable print size
Internet Archive	~870	2–3 px	Warm cast, correctable	Up to 30 x 40 in
HathiTrust	~690	2 px	Accurate, calibrated	Up to 24 x 36 in
British Library	~500 (soft)	5–6 px	Best of the four	Up to 12 x 18 in
Google Books	~300	6+ px with JPEG halo	Pink cast, crushed	Up to 8 x 10 in only

What This Does NOT Prove

This is a single-title audit. Four scans of one book, measured on one plate with a specific letterform. It does not prove that HathiTrust's scans are always better than the British Library's, or that Google Books is always the worst option. Institutional scanning workflows change over time, and any given plate can be an outlier — a page that was rescanned after damage, a title that was digitised on a higher-end rig for a specific exhibition. Any designer using an archive scan should run the same edge-transition test on the specific file they intend to work from.

It also does not prove anything about the underlying 1886 edition. All four scans are working from physical copies of varying condition, and we cannot separate what the scanner did badly from what the source page had already suffered. The Internet Archive's warm cast may be a real feature of a paper stock that has yellowed unevenly. Colour fidelity, in a strict sense, is fidelity to a book that no longer exists in the state Kate Greenaway saw it in.

The Takeaway

Trust the pixels, not the metadata. Run an edge-transition test on any archive scan before you build a wall-scale reproduction on top of it — the DPI number in the file properties will lie to you about half the time.

FAQ

Why does edge transition matter more than DPI?

Because DPI reports pixels per inch across the page, and pixels per inch across the page is not the same as sharpness at the letter. A scan can be upsampled from a smaller original — bicubic interpolation adds pixels without adding information. The letter looks the right size on screen but resolves over a wide, soft edge instead of a crisp one. Measuring the black-to-paper transition in pixels tells you the acuity the file actually carries, which is what determines whether the print will hold up at nursery scale.

What size print does 600 DPI at the letter actually support?

A file that resolves 600 DPI at the letterform and holds a two-to-three-pixel edge transition will print cleanly at up to roughly 30 by 40 inches when the viewing distance is over two feet. Below 300 DPI at the letter, the same file starts to show visible pixel steps in the serifs at any wall-scale size. The letter itself is the reference — not the page dimension, which can be padded with margin.

Can I colour-correct the Internet Archive scan to match the British Library scan?

Partially. You can neutralise the paper base by sampling a plain paper area and applying a global white-point adjustment. What you cannot easily reconstruct is the chromolithograph colour that was compressed toward the warm end of the range. Once channel information is lost to a colour cast, matching a properly calibrated scan is estimation, not correction. If colour fidelity matters, work from the British Library file and use the Internet Archive only for linework.

Why did A Apple Pie become a reference point for nursery alphabet design?

Kate Greenaway's 1886 edition is one of the earliest alphabet books that treats each letter as an illustrated set piece rather than a decorated capital. That per-letter treatment maps directly onto how a wall-scale alphabet chart is composed today, with each letter carrying its own vignette. The historical record is the reason the book keeps getting scanned, and the reason a nursery studio comparing our own alphabet charts against period sources ends up at this file in the first place.

Do the tests transfer to other Kate Greenaway books scanned by the same archives?

Partly, and only as a starting hypothesis. The Internet Archive tends to prioritise pixel density and neglects colour calibration across its Greenaway holdings, and HathiTrust tends to run neutral grey-card calibration consistently. But scan quality is title-by-title, sometimes plate-by-plate. Under the Window and Marigold Garden should be tested individually if a designer intends to build a print from them. The measurement is fast — one plate, one letter, one edge transition — and it saves a print run.

What print resolution does a nursery-scale alphabet chart actually need?

Between 300 and 400 DPI at the finished print size is the working threshold for a chart viewed from four feet away. Below 300 the letterforms start to soften at reading distance for a small child. Above 400 the returns diminish. For our own studio work we target 400 DPI at 24 by 36 inches, which requires a source file with roughly 600 DPI at the letter to survive the retouching and cropping pipeline without hitting the floor.

Is there a difference between scanning DPI and printing DPI?

Yes, and confusing them is the fastest way to a disappointing print. Scanning DPI describes how densely the scanner samples the physical page. Printing DPI describes how densely the printer lays ink on the finished sheet. A file scanned at 600 DPI does not automatically print at 600 DPI — if you scale the image up for a larger output, effective printing DPI drops. Sizing math has to be done at the output stage, not assumed from the archive metadata.