Digital fax OCR in healthcare reads text from black-and-white fax images, usually TIFF or PDF files, and works well on typed pages sent at fine resolution. Results drop on standard-resolution pages, skewed or speckled images, small print and handwriting. Getting better results means cleaning the image first and sending doubtful fields, such as dates, digits and doses, to a person.
What does a digital fax file look like to OCR?
A digital fax file, to OCR software, is a stack of black-and-white page images with no text inside. Each pixel is black or white, never gray. The file is usually a TIFF, sometimes a PDF wrapping the same images, and each page was sent at standard or fine resolution, which sets how much detail OCR gets.
OCR, optical character recognition, is the step that turns the shapes on a page image into characters. On a fax it has no text layer to fall back on, so everything it returns comes from the pixels.
The usual fax file format has its own published profile. RFC 2306, the TIFF-F profile for facsimile, describes TIFF-F as a file format for storing and exchanging fax images, and it sets bits per sample to 1, because "TIFF-F is only used for black-and-white facsimile images." There is no gray to smooth the edges of letters, so a stroke that falls between two pixels becomes either a full pixel or nothing.
The same RFC lists the resolutions a fax image can carry. Horizontal resolution is typically 204 dots per inch. Vertical resolution is what separates the two common modes: the RFC says "almost all facsimile products support both standard (98 dpi) vertical resolution and 'fine' (196 dpi) resolution." A standard page has about half the rows of a fine page.
| Feature | Standard resolution | Fine resolution |
|---|---|---|
| Pixels per inch, horizontal by vertical | 204 by 98 | 204 by 196 |
| Small print, such as footnotes and reference ranges | Letters lose their middle strokes and can merge | Usually readable, though thin fonts still break up |
| Digits | 3, 5, 6, 8 and 9 can lose the gaps that tell them apart | Gaps between strokes mostly survive |
| Decimal points and slashes | Can vanish or turn into specks | More often kept |
| Handwriting | Thin pen strokes break into dashes | Strokes hold together better, still hard for OCR |
| Typed body text at normal size | Often readable | Readable |
The table describes typical effects, not measured rates. Results vary with the font, the original paper and how many times the page was copied before it was faxed.
When a fax arrives as a PDF, check whether it holds any text before you plan an OCR step. A PDF made from a fax normally holds page images only, so selecting text returns nothing until OCR runs.
OCR for fax TIFF files: multi-page and compression quirks
OCR for fax TIFF files depends on the software reading the file correctly before it reads any letters. Fax TIFFs pack every page into one file and compress each page with fax-specific codes, Group 3 or Group 4. A reader that handles only part of that structure can skip pages or fail to open the file at all.
Multi-page structure. RFC 2306 explains that in a fax TIFF "each facsimile page in a multi-page document has its own IFD", an image file directory, and the directories form a linked list, each pointing to the next. A tool that reads only the first directory returns only page 1. When OCR output for a 6-page fax holds one page of text, check this before blaming image quality.
Page numbers inside the file. The same RFC defines a PageNumber field holding the page number and the total pages, and notes that a total of 0 means the total "is not available." Comparing that total with the number of pages that actually came through OCR is a cheap completeness check.
Compression codes. The Group 3 standard is ITU-T Recommendation T.4, standardization of Group 3 facsimile terminals for document transmission. RFC 2306 allows two compression values in a fax TIFF: 3, for T.4 Modified Huffman or two-dimensional Modified READ coding, and 4, for T.6 Modified Modified READ coding. The Library of Congress format description for TIFF bitmap with Group 4 compression notes that this format is "limited to bitonal imaging" and has been widely used for scanning business documents into document management systems.
Reader support. RFC 2306 warns that "some historic TIFF-F readers have not been able to read" files using two-dimensional coding, and recommends that readers support all three methods. It also says readers must handle both bit orders, since most fax products store data least significant bit first. An older converter that gets either detail wrong produces a blank page, a page of noise or an error.
The practical rule for a healthcare team: before judging OCR quality, confirm that every page of every fax opened, in the right order, at the right size.
Why does OCR accuracy drop on low resolution faxes?
OCR accuracy drops on low resolution faxes because standard mode captures only about half the rows of fine mode, so thin strokes, small gaps and punctuation go missing before OCR starts. Skew, speckle and copies of copies make it worse. Each problem removes or adds pixels that OCR then reads as part of a letter.
Standard mode. At 98 rows per inch, a small letter "e" has only a few rows to show its crossbar and opening. When the opening fills in, "e" becomes "c" or "o". The open tops of 4 and 9, and the gaps in 3 and 8, close up the same way.
Below the size OCR engines prefer. The Tesseract OCR documentation on improving quality, for one open source engine, says "Tesseract works best on images which have a DPI of at least 300 dpi." Both fax modes sit under that vertically. Rescaling a fax image up does not add detail that was never transmitted, but it can help an engine that expects larger characters.
Skew. A page fed into the sending machine at an angle arrives with slanted text lines. Slanted lines are harder for an engine to follow, which is why straightening the page comes first in the cleanup steps below.
Speckle. Dust on the sender's glass, paper texture and line noise add isolated black dots. A dot beside a 1 can turn it into an i, and a dot between digits can look like a decimal point.
Copies of copies. Many faxed pages were printed from a fax, annotated, and faxed again. Each round thickens some strokes and drops others. By the third generation, small print can be unreadable to people and software alike.
The general causes of poor image quality, outside fax, are covered in how OCR handles low quality medical scans. This section covers the parts specific to fax transmission.
Cleaning fax images before OCR
Cleaning fax images before OCR means fixing problems in the page image so the OCR engine sees straighter, cleaner text: deskewing slanted pages, removing speckle, rotating pages that arrived sideways or upside down, and trimming dark borders. Each step helps in the right case and can damage the page when applied blindly.
The Tesseract documentation describes these steps and their purpose:
- Deskewing. It defines a skewed image as one "scanned when not straight" and advises rotating the page "so that the text lines are horizontal."
- Noise removal. It describes noise as "random variation of brightness or colour in an image, that can make the text of the image more difficult to read", and notes that some noise survives the engine's own binarisation step.
- Borders. It notes that "scanned pages often have dark borders around them", which can be read as characters.
- Built-in processing. It explains that the engine does various image processing operations internally before OCR, but that this "can result in a significant reduction in accuracy" when it is not good enough.
Rotation matters more on faxes than on scans. Senders feed pages face up, face down and sideways. If your OCR tool can detect page orientation and rotate before reading, turn that on. Check that the rotation step runs on every page of a multi-page file, not only the first.
When cleanup hurts. Fax images are already black and white, so there is little margin. A despeckle filter tuned for scans can erase decimal points, the dots on i and j, and the thin strokes of a slash in a date. A strong deskew on a page with a skewed stamp, but straight text, can tilt the text instead. Test each step on a set of your own fax pages, de-identified, and compare the output with and without it before turning it on for everything.
Keep the original image. Cleaning makes a working copy for OCR. The page that goes into the chart, and the page a reviewer checks against, should be the one the sender transmitted.
Fax OCR fields that need a second look
Fax OCR fields that need a second look are the ones where a small misread changes the meaning: handwriting, digits that fax compression blurs together such as 1 and 7, dates, and drug doses with decimal points. These fields should go to a person who checks the value against the page image before it is used anywhere.
Handwriting. Signatures, initials, margin notes and handwritten orders are the hardest content on a fax. Fax resolution breaks thin pen strokes, and handwriting is harder for OCR than typed text even on a clean scan. Treat handwritten values as unread until a person confirms them.
Digits. On a standard-mode page, 1 and 7, 3 and 8, 5 and 6, 0 and 8, and the letter O and zero are frequent confusions. In a member ID, a medical record number or a phone number, one wrong digit points to a different person.
Dates. A date like 01/07/2026 can become 07/07/2026 or 01/01/2026 with one digit wrong. Slashes can vanish, turning 1/12/26 into a string that parses as a different date. Birth dates and dates of service both drive matching and billing.
Drug doses and quantities. A decimal point that drops out turns 2.5 into 25. A trailing zero that should not be there, such as 5.0, can be read as 50 when the point disappears. These are data entry checks: the reviewer confirms that the extracted characters match the page, and any question about the dose itself goes to the clinician.
A useful check is to have OCR return a confidence score for each field and route the low ones to review. How those scores work, and how to set the cut-off, is covered across the posts on confidence scores and human review. The fictional EOB in the Schema Care threshold demo shows the same idea on a remittance: raise the threshold and more fields wait for a person instead of passing straight through.
When fax OCR is the wrong tool
Fax OCR is the wrong tool when the page is too degraded for anyone to read, when the content is mostly handwritten, or when the sender could send the same data electronically. In those cases, asking for a resend at fine resolution or for an electronic copy costs less than correcting OCR output field by field.
Ask the sender for a resend when:
- A key identifier, such as name, birth date or member ID, cannot be read by a person looking at the image.
- Pages are missing, cut off at the bottom, or the PageNumber total does not match what arrived.
- The page was clearly sent in standard mode and holds small print, such as a lab report's reference ranges.
Ask for an electronic copy instead when the same sender faxes you the same kind of document every day. A hospital or reference lab that sends results by fax may also be able to send them through an interface or a portal. The posts on fax to EHR automation and on what healthcare fax automation costs and saves cover how to weigh those routes against improving fax handling.
Some pages are better read by a person from the start. A two-line handwritten referral note does not need an OCR step; a staff member can enter it faster than they can check OCR output for it.