Grayscale vs Black and White Scanning for OCR
By Schema LawPublished Updated
For OCR, choose grayscale scanning unless every page is crisp black print on clean white paper. Black and white scanning decides at the scanner which dots count as ink and discards everything else, so faint text is gone before any software sees it. Scan in color when color itself carries meaning on the page.
The choice is made once, at the scanner, and it cannot be reversed afterwards. A grayscale file can be turned into a black and white one at any time. A black and white file cannot be turned back, because the shades it would need were never saved.
Should you scan in grayscale or black and white for OCR?
Scan in grayscale for OCR as your default. Use black and white only for clean, dark, laser-printed originals where small files matter more than a safety margin. Grayscale hands the recognition software the full picture and lets it decide what is ink. Black and white makes that decision in the scanner, with less information.
The names on the control panel vary. Black and white may appear as "text", "bitonal", "1-bit", "line art" or "B&W". Grayscale may appear as "gray", "8-bit" or, on some machines, "monochrome", a word other machines use for black and white. If a label is unclear, scan a page with a pencil note on it and look at the result. Pencil that shows as soft gray means grayscale. Pencil that is missing or broken into black specks means black and white.
A recognition engine may itself convert a gray image to black and white before it reads the letters. That sounds like an argument for doing it at the scanner, but it is the opposite. The documentation for the open-source engine Tesseract, on improving the quality of its output, says the engine does this conversion itself and that the result can be suboptimal when the page background is of uneven darkness. It also notes that version 5 added adaptive methods for the step. Software can only apply a smarter method if it receives the grays.
Institutional practice leans the same way. The federal regulation on digitizing permanent paper records, 36 CFR 1236.50, lists the color mode for modern textual paper records as color or grayscale. That rule applies to federal agencies, not to law firms, but it is a useful reference when you write a scanning standard.
Color mode works together with resolution. The post on the best dpi for OCR explains how many dots per inch a page needs, and neither setting can make up for a poor choice in the other.
What does black and white scanning do to each pixel?
Black and white scanning measures how bright each pixel is and compares it with a cutoff, called the threshold. Darker than the cutoff becomes black. Lighter becomes white. One bit is stored per pixel, and the measured brightness is thrown away.
In a grayscale scan each pixel is a number, commonly from 0 for black to 255 for white. Take a short run of pixels across the edge of a printed letter and into a pencil mark beside it:
| Pixel | Grayscale value | What it is | Black and white at a cutoff of 128 |
|---|---|---|---|
| 1 | 245 | Paper | White |
| 2 | 190 | Soft edge of a printed stroke | White |
| 3 | 40 | Printed stroke | Black |
| 4 | 110 | Soft edge on the other side | Black |
| 5 | 240 | Paper | White |
| 6 | 150 | Pencil line | White |
| 7 | 235 | Paper | White |
Three things happened in that row. The printed stroke survived. Its soft edges were pushed one way or the other, which is why letters in a black and white scan have ragged outlines. And the pencil line, at 150, disappeared completely.
One cutoff for a whole page
Many scanners apply a single cutoff to the entire sheet. That works when ink is uniformly dark and paper uniformly light. It fails when the page is uneven:
- A photocopy with a gray cast has a background near the cutoff, so it turns into a field of black speckle.
- A shadow in the gutter of a bound volume goes solid black and swallows the words in it.
- Thin strokes, such as the hairlines of a serif face, fall on the light side and break, so "h" reads as "li" and "m" as "rn".
- Thick or smudged strokes fill in, so "e" and "o" both become blobs.
The "darkness" or "density" dial on a scanner moves the cutoff. Turning it up rescues faint text and blackens stains. Turning it down cleans the background and erases the faint text. No position is right for a page that has both.
Dithering is not grayscale
Some scanners offer a "photo" or "halftone" option inside black and white mode. It imitates gray by scattering black dots in patterns. That looks acceptable for a photograph and is harmful for text, because the letters are now built from dot patterns. If you want shades, pick true grayscale.
What grayscale keeps: pencil, faint stamps and highlighter
Grayscale keeps every mark that is lighter than solid ink: pencil, pale stamp impressions, highlighter, colored pen, and the shading of worn or copied pages. In black and white, each of those either drops to white and vanishes or rises to black and covers what is under it.
What happens to common marks in a law office file:
| Mark on the paper | In black and white | In grayscale |
|---|---|---|
| Pencil annotation in a margin | Often gone, or a few specks | A gray line you can read |
| Faint date or "RECEIVED" stamp | Partly or wholly missing | Pale but present |
| Yellow highlighter over text | Usually drops out; the text stays | A light band with dark text on it |
| Pink, blue, green or orange highlighter | May turn into a solid black bar over the text | A mid-gray band; the text is darker and still readable |
| Blue ballpoint signature | Thin, broken strokes | Complete strokes in gray |
| Carbon copy or faded typewriting | Letters break up or disappear | Weak but continuous letters |
| Colored paper, such as a pink or goldenrod copy | Heavy speckle or a black page | An even gray background |
| Show-through from the reverse side | Random black fragments | A faint ghost the software can discount |
| Embossed seal | Usually invisible | Sometimes visible as slight shading |
The highlighter row deserves attention. A reviewer highlights the sentences that matter. If a dark highlighter becomes a black bar, the scan has removed exactly the passages someone flagged, and the gap is easy to miss because the rest of the page looks fine.
Grayscale earns its place on a highlighted transcript. The marker stays light enough that you can still read the line under it. Check one page for the name of a party, the hearing date, and a statute the highlighter crossed. A black and white pass can turn that mark into a solid bar and cover the fact you meant to find.
Stamps raise a related problem. A stamp that lands on top of typed text hides part of it in any mode, but grayscale keeps the stamp lighter than the print so the two can be told apart. The post on whether OCR can read stamps over text goes through which stamps cause trouble and what to do about them.
Grayscale also protects you against things you did not notice at the scanner. Nobody inspects every sheet in a box before feeding it. A default that keeps faint marks means the one page with a pencilled figure in the margin is captured without anyone having to spot it first.
When does color scanning earn its larger file?
Color scanning earns its larger file when the color of a mark tells the reader something that its darkness does not. For recognizing black text, color rarely beats grayscale. For preserving what a page means, it sometimes matters a great deal.
Cases where color is worth it:
- Color-coded review. Exhibits highlighted in different colors by issue or by reviewer. In gray, the colors merge into similar bands.
- Redlines. Printed comparisons where insertions and deletions are shown in red and blue.
- Ink that shows an original. A blue signature or a colored notary stamp can help show that a page is an original and not a photocopy.
- Colored stamps over black text. A red stamp and black print can be close in darkness and far apart in hue. Color keeps them separable for a person, and for software that uses color.
- Photographs, maps, plats and charts. A legend that says "the red line is the boundary" is meaningless in gray.
- Medical and scientific printouts. Charts and images where color encodes a value.
- Handwritten corrections in several inks. Who wrote what, and in what order, may depend on telling the inks apart.
Federal records practice draws the line in the same place. The regulation on digitizing permanent paper records requires agencies to use an RGB color mode when records contain color that is necessary to interpret the information in them.
Two cautions. First, if the scan will be produced or filed, check the governing requirements before you choose. Read the production specification and your court's filing rules for any statement about color or scanning settings, because they vary. The post on searchable PDFs for court filing covers how to find your court's rule. Second, a scanner's "auto color" feature decides page by page whether color is present. It saves space, but it can class a page with one small colored mark as black and white. Spot-check its decisions on a sample before you trust it on a whole box.
How color mode changes the size of a scanned file
Color mode sets how much data each pixel holds before compression: one bit in black and white, eight in grayscale, twenty-four in color. On the same page at the same resolution, raw grayscale is eight times the size of raw black and white, and raw color is three times the size of grayscale.
The arithmetic for one letter-size page at 300 dpi, which is 2,550 by 3,300 pixels, or 8,415,000 pixels:
| Mode | Bits per pixel | Raw size of one page |
|---|---|---|
| Black and white | 1 | About 1 MB |
| Grayscale | 8 | About 8.4 MB |
| Color | 24 | About 25 MB |
Nobody stores raw pages. Compression shrinks all three, and it shrinks them unevenly:
- Black and white compresses very well without loss. A text page is mostly long runs of white, which the compression methods designed for fax and bitonal images pack tightly.
- Grayscale and color are usually compressed with loss. JPEG-style compression discards fine variation. The quality setting matters more than the mode. At a high quality setting the page is faithful and large. At a low one, smudges appear around every letter.
That second point is a trap. A grayscale scan saved at the scanner's "smallest file" setting can read worse than a clean black and white scan, because the compression has damaged the letter edges. If you choose grayscale, choose moderate compression with it.
Measure your own scanner
Compressed sizes depend on the device, the paper and the settings, so measure them. Scan the same ten typical pages three times, once in each mode, at the resolution you plan to use. Divide each file size by ten. Now you have real per-page figures.
Then do the planning sum. Suppose, purely as an illustration, your figures came out at 60 KB a page in black and white, 400 KB in grayscale and 900 KB in color. Against a 20 MB limit, which is the largest upload Schema Law accepts, that is about 330 pages, 50 pages and 22 pages. Your numbers will differ. The point is that the mode you choose decides where a long document has to be split, and it is better to know that before scanning a 300-page exhibit than after.
A mixed approach keeps files manageable. Scan the bulk of a file in grayscale. Pull the pages where color matters and scan those in color. Keep both in the same order as the paper, and note in your scanning log which pages were done in which mode.
Choosing grayscale, color or black and white by document type
Choose by what is on the paper, and by what you would lose if a faint mark disappeared. When you cannot inspect every page, pick the mode that suits the worst page in the batch, not the best.
| Document | Mode | Why |
|---|---|---|
| Fresh laser-printed pleadings, letters and contracts | Black and white is acceptable; grayscale is safer | Dark, even print survives a cutoff |
| Photocopies, especially copies of copies | Grayscale | Gray backgrounds and weak strokes need shades |
| Client files with handwritten notes | Grayscale | Pencil and light ink must survive |
| Faxed pages printed on paper | Grayscale | Broken strokes are better preserved |
| Carbon copies, typewritten originals, faded pages | Grayscale | Low contrast between ink and paper |
| Pages on colored paper | Grayscale | Avoids speckle and black pages |
| Signed originals where ink color may matter | Color | Shows blue ink and colored seals |
| Exhibits highlighted or tabbed in several colors | Color | The coding is part of the content |
| Redlined drafts and color charts | Color | Meaning depends on hue |
| Photographs, plats, maps and plans | Color, or grayscale for black and white originals | Tonal detail is the content |
| Receipts on thermal paper | Grayscale | Print is pale and fades unevenly |
| A mixed box nobody has reviewed page by page | Grayscale, with color for pages pulled by hand | Protects the pages you did not look at |
A few questions settle most remaining cases:
- Will the paper be destroyed or returned after scanning? If so, lean toward the richer mode. You cannot go back.
- Is there any handwriting, stamp or highlighter in the set? If you do not know, assume yes.
- Does a size limit apply downstream? If an e-filing or upload limit is tight, grayscale with sensible compression and a split at document boundaries is usually a better answer than dropping to black and white.
- Has someone told you what they need? A court, a client or the receiving party may have stated a preference. Follow it, and keep a richer master for yourself if the paper will not survive.
Color mode is one line in a longer list of settings. The guide to digitizing a firm's paper has the scanner settings that make OCR work, including duplex, blank-page removal and file format.
Whatever you choose, test it on real pages before committing a whole project to it. Scan one difficult page in each mode, run each through your OCR tool and compare the text with the paper. Schema Law's free plan reads one document a day of up to three pages, which is enough to put the three versions of a single page side by side. The export formats include plain text and Word, either of which makes that comparison quick.