Preserve the tonal evidence the document actually contains
Use color when color carries meaning, grayscale for faded or uneven pages, and bitonal only when the page is clean enough that thresholding will not erase meaningful marks.
You should finish with a richer source master and a tested recognition derivative, chosen because meaningful strokes and annotations survive rather than because the background looks cleaner.
Apply the decision with Image to Text. Then test the tradeoff against Best Scan DPI for OCR: Choose Resolution by Character Size and Historical Document OCR: Type, Damage and Layout.
If the source arrives as a compressed web asset, test it directly with the WebP text extraction tool before converting or discarding color information.
Scan mode decision
| Source | Starting mode | Reason |
|---|---|---|
| Clean black print on white paper | Bitonal or grayscale test | High contrast may tolerate thresholding |
| Faded carbon copy | Grayscale | Tonal differences preserve weak strokes |
| Colored annotations or stamps | Color | Hue is part of the evidence |
| Mixed photographs and text | Color or grayscale | Continuous tone matters beyond OCR |
Keep a rich source and test simpler derivatives
For a faded form with a blue stamp and pencil annotation, color preserves evidence that grayscale or bitonal output may merge or erase. Keep the color capture as the source. Create separate derivatives only for recognition tests, and name the settings used for each.
Inspect weak strokes, punctuation, stains, show-through, highlights, and colored marks. A bitonal page may look clean while losing the faintest entries. Choose the derivative that reduces meaningful OCR errors without discarding marks required for interpretation, and never replace the richer source with the simplified copy.
- Preserve color when color carries meaning
- Treat thresholding as reversible workflow data
- Compare consequential regions
Choose a scan mode by the marks that must survive
Start with what must be preserved
Colored annotations, stamps, highlights, and inks can change interpretation, so capture them in color. Grayscale retains tonal differences without hue. Bitonal reduces every pixel to black or white and can permanently remove faint strokes.
Treat thresholding as a derivative
Binarization can make clean print easier to segment, especially with a controlled background. On stained, folded, or faded pages, one threshold may erase light text while turning shadows into false characters. Keep the richer source for another attempt.
Match mode to the original
National Archives guidance distinguishes clean high-contrast printed material from low-contrast, stained, faded, illustrated, or annotated records. That distinction is more useful than applying one scan mode to an entire mixed collection.
Compare consequential regions
Test faint pencil, small punctuation, stamps, reverse-side show-through, and colored notes. Choose the derivative with the fewest meaningful errors, not the one with the whitest background.
Scan-mode selection checklist
- Identify whether color carries meaning
- Retain a color or grayscale master
- Create bitonal output only as a derivative
- Check faint marks after thresholding
- Document the chosen mode and settings
Cleaner is not always more faithful
Thresholding may erase faint marks, color-coded meaning, stamps, highlights, or paper evidence. Keep a richer master even when a simplified derivative recognizes text better.
What the imaging sources establish
The references explain image modes and preservation principles. A representative recognition comparison determines which derivative is useful without discarding the source master.