"Convert this to black and white" means three completely different things depending on who is asking, and picking the wrong one produces either a file nobody's system will accept or a 40 MB scan of a two-page form.
The variable is bit depth: how many bits are spent describing each pixel.
The three options
24-bit colour spends 8 bits each on red, green and blue: 16.7 million possible values per pixel. This is what a phone camera and a colour scanner produce by default.
8-bit grayscale spends 8 bits on a single brightness value: 256 shades from black to white. A photograph still looks like a photograph. This is what most people mean by "black and white" in casual speech, and it is what almost every online "PDF to black and white" tool actually produces.
1-bit — also called bitonal or monochrome — spends a single bit per pixel. Ink or paper. Nothing in between. This is what fax machines, cheque imaging systems, microfilm archives and most document-management systems mean when they specify black and white, and it is the one people usually cannot find a tool for.
Note that "monochrome" is the loosest of these words. It means one colour, so a grayscale scan is technically monochrome too. When a specification says monochrome and also says 200 or 300 DPI, it almost always means 1-bit.
The size arithmetic
The ratios are exactly the bit depths, before any compression enters the picture.
A US Letter page at 300 DPI is 2550 × 3300 pixels, which is 8,415,000 pixels:
| Bit depth | Raw page data | | --- | --- | | 24-bit colour | 25.2 MB | | 8-bit grayscale | 8.4 MB | | 1-bit bitonal | 1.05 MB |
One eighth of grayscale. One twenty-fourth of colour. And that is before compression, which is where bitonal pulls further ahead: a scanned text page is mostly long runs of identical white pixels, and both Flate and CCITT Group 4 collapse those runs to almost nothing. Real scanned text pages routinely land between 20 KB and 80 KB per page as 1-bit, against several hundred KB as grayscale JPEG.
For a 200-page scanned manual, that is the difference between a file you can email and a file you cannot.
What CCITT Group 4 has to do with it
Group 4 (also written G4, or ITU-T T.6) is the run-length compression scheme fax machines have used since the 1980s. It encodes runs of black and white along each scan line, using the line above as a reference. It only works on 1-bit data — that is the whole design — which is why "1-bit", "bitonal" and "fax quality" get used as though they were the same word.
Group 4 is usually the best compressor for clean scanned text. Flate — the zlib deflate used throughout PDF — is often better for dithered images, where the pixel pattern is noisy and defeats run-length encoding.
Our PDF to Black & White tool produces genuine 1-bit pixel data (a DeviceGray image XObject with /BitsPerComponent 1) compressed with Flate, because Flate is available natively in the browser and a Group 4 encoder is not. If you specifically need Group 4 — some archival specifications require it — render locally with Ghostscript:
gs -sDEVICE=tiffg4 -r300 -dNOPAUSE -dBATCH -sOutputFile=scan.tif scan.pdfThreshold versus dithering
Reducing 256 shades to 2 means throwing away 254 of them. There are two philosophies about how.
A hard threshold picks a brightness cutoff. Everything above it becomes paper, everything below becomes ink. For text this is exactly right: letterforms are already near-black on near-white, and thresholding sharpens them. For a photograph it is a disaster — a face becomes flat abstract patches, because every tone in the mid range collapses to one side or the other.
Dithering simulates intermediate tones by scattering black and white pixels in a pattern your eye blends at normal viewing distance. Three that matter:
- Floyd-Steinberg pushes each pixel's quantisation error into its not-yet-processed neighbours, weighted 7/16 right, 3/16 down-left, 5/16 down, 1/16 down-right. It gives the most natural-looking grain and is the default choice for photographs.
- Atkinson, developed at Apple for the original Macintosh, distributes only 6/8 of the error rather than all of it. Losing a quarter of the error deliberately means highlights stay bright and shadows stay dark, so it holds contrast far better on faded or unevenly lit scans — at the cost of losing some detail at the extremes.
- Ordered (Bayer) dithering compares each pixel against a fixed matrix rather than diffusing anything. The result is a regular cross-hatch. It is the fastest, it is completely deterministic, and some print and imaging workflows prefer it precisely because the pattern is predictable.
The rule: threshold for anything made of type and lines, dither for anything made of tone.
Choosing a threshold
If you are thresholding, the cutoff is the only setting that matters, and the right value depends on the scan rather than on theory.
- Text is breaking up or disappearing: raise the threshold. More mid-tones become ink, so faint pencil, worn print and low-contrast scans come back.
- The page background is turning grey or black: lower the threshold. Yellowed paper, shadows near the spine and scanner backlight bleed all sit in the mid range and can flip to ink at a default cutoff.
Do one page first and look at it before committing a 200-page document.
Resolution matters more at 1 bit
At 8 bits per pixel, a soft edge is represented by a gradient of grey. At 1 bit there is no gradient — an edge is either on this pixel or the next one. That makes bitonal output much more sensitive to resolution than grayscale.
- 150 DPI: acceptable for large type, poor for anything under 10 point.
- 200 DPI: the fax "fine mode" standard, and the practical floor for ordinary document text.
- 300 DPI: the archival norm, and the threshold below which OCR accuracy starts falling off noticeably.
- 400 DPI and above: worth it only for small print, dense tables, or engineering drawings with fine hatching.
Because the pixel data is only one bit deep, going from 200 to 300 DPI costs far less than the same step would at 8 or 24 bits. Scanning a text document at 300 DPI bitonal is still smaller than 150 DPI grayscale.
Print cost, briefly
Colour pages are billed at colour rates by almost every print shop and managed-print contract, and the trigger is usually any non-grey ink anywhere on the page — a blue hyperlink in a footer is enough. Converting to grayscale before sending removes the ambiguity, because there is no colour left to detect. That is a different job from 1-bit conversion, and for anything containing photographs it is the one you want.
A decision table
| Source | Want | Use | | --- | --- | --- | | Scanned text, forms, contracts | Smallest file, crisp type | 1-bit, hard threshold, 300 DPI | | Scanned pages with photos | Legible photos, small file | 1-bit, Floyd-Steinberg, 300 DPI | | Faded or unevenly lit scan | Readable output | 1-bit, Atkinson, raised threshold | | Colour report going to print | No colour billing | Grayscale | | Anything where text must stay selectable | Searchable output | Neither — see below |
The catch worth knowing before you start
Both grayscale and 1-bit conversion here work by rendering each page to an image and rebuilding the document from those images. The page looks the same, but selectable text, links, form fields and accessibility tags do not survive. That is inherent to changing every pixel on the page, not a shortcut.
If you need reduced colour *and* selectable text, convert the colour space instead of rasterizing:
gs -sDEVICE=pdfwrite -sColorConversionStrategy=Gray \
-dProcessColorModel=/DeviceGray -dNOPAUSE -dBATCH \
-sOutputFile=gray.pdf colour.pdfGhostscript rewrites the colour operators and leaves the text layer alone. There is no browser equivalent, and pretending otherwise would not make one exist.
Related tools: PDF to Black & White for true 1-bit output with all four dithering modes, PDF to Grayscale for tonal output, and Compress PDF when the images rather than the bit depth are the problem.