How to OCR a Faxed Document and Trust the Result

To OCR a faxed document, start from the file your fax service saved, not a printout you scanned again. Upload it, export text or Word, then proofread the digits, names and small print against the fax image. Fax resolution is low enough that those characters are the first to go wrong.

Faxes still turn up throughout legal work: medical records, insurance correspondence, signed authorizations, notices from agencies and courts that have not moved on. The steps below get usable text out of them and tell you which parts of that text deserve suspicion.

Why are faxes hard for OCR?

Faxes are hard for OCR because the fax standard sends pages at low resolution in pure black and white. Each letter arrives built from very few dots, with no gray to soften its edges, and whatever dirt the sending machine had on its glass is sent along with the text.

The resolution has known numbers. A standard-mode fax records about 204 dots per inch across the page and only about 98 lines per inch down it, and the mode most machines label "fine" doubles the lines. The post on low resolution scans gives the fax resolution figures and what each mode leaves of a letter.

Resolution is only the start. The rest of the damage comes from how fax machines capture and send a page, and all of it happens before the file reaches you:

• No gray. Every dot is black or white. Light text drops out, shaded boxes become a field of speckle, and photographs or letterhead logos turn into dark blotches.
• A faint original. The sending machine decides what counts as ink. Pencil, a pale carbonless copy or a signature date in light blue pen can fall on the white side of that decision and never be sent at all.
• Dirt from the sender's glass. Specks are transmitted as black dots. A speck beside a word reads as a period or a comma, a speck after a digit looks like a decimal point, and a speck in the opening of a "c" turns it into an "e" or an "o".
• Line faults. A noisy phone line can drop or repeat a row of dots, which shows as a thin streak or a sliced line of text. A dirty sending scanner draws a vertical black line down every page.
• A crooked feed. A sheet pulled through the sender's feeder at a slight angle arrives with sloping lines of text, and coarse dots turn each sloping stroke into a staircase. The post on how to OCR a crooked scan covers tilted pages in general.
• Copies of copies. A fax that was printed and faxed on is captured a second time on a new grid of dots. The jagged edges from the first trip are roughened again, and thin strokes that survived once often do not survive twice.

None of these can be repaired at your end, so look for them on the image before you convert. Take a fictional two-page denial letter faxed by an insurer, Castlereach Mutual. The typed body in 12 point arrives readable. The claim number sits in a shaded box and comes through as speckle. The adjuster's handwritten date beside the signature, in light blue ink, is missing altogether. A converter will return clean paragraphs for that letter and say nothing about the two facts you most needed, which is why the image gets a look first.

Picture a faxed order instead of that denial letter. You still read the motion or pleading title, the summons date if one is on the face, the docket number, and any line about service of process. Those sit in small type or in a header the fax chewed, and a later reader will copy them into a calendar. Check them on the image.

Converting a fax to text or Word

Converting a fax to text or Word takes four steps: get the best digital copy, upload it, export the format you need, and check the result against the image. The first step does the most good, and it is the one people skip.

1. Find the original file. If the fax arrived through a fax server, an online fax service or a multifunction copier, it already exists as a PDF or TIFF. Use that file. Printing the fax and scanning the printout adds a second round of blur and skew to an image that had no quality to spare.
2. Convert TIFF if you have to. Some fax systems save multi-page TIFF files. Schema Law takes PDF, PNG or JPEG and not TIFF, so save the TIFF as a PDF first. Use a "save as PDF" or "print to PDF" route that keeps the image at its full size, and avoid any option that reduces file size.
3. Leave the image alone. Do not sharpen it, increase contrast or crop it in a photo editor unless you have tested that the change helps. Enlarging a fax in an editor cannot add detail that was never transmitted, and the post on low resolution scans explains whether enlarging a low-resolution image helps recognition.
4. Upload and wait for the pages to be read. In Schema Law you sign in, drop the file on the upload area and view the parsed pages when they are ready.
5. Export for the job. Choose plain text for fax to text, where you want words to search or paste. Choose Word for fax to Word, where you will edit. Choose searchable PDF when the fax itself is the record and you only need to find things in it.
6. Proofread before you use it. The section on which characters to check first gives the order.

Keep the fax file itself, unchanged, next to whatever you export. The export is a working copy. The fax is what was received, complete with the header that shows when and from where.

If the fax is headed for editing, the Word file will need the same review any converted scan does. The guide to converting a scanned PDF to Word covers that cleanup, so it is not repeated here.

Does OCR work on fax-quality scans?

OCR works on fax-quality scans well enough to make them searchable and to save retyping, provided the text is ordinary body type and the transmission was clean. It works poorly on small print, forms with shaded fields, handwriting and any fax that was itself a copy of a fax.

A rough guide to what to expect:

The mode the sender's machine used makes the largest single difference to a typed page, and the post on low resolution scans shows how to tell standard from fine fax mode from the file itself.

No general accuracy figure applies to your fax. Two faxes of the same letter from two machines can differ more than two different letters from the same machine. The practical test is quick: convert one typical page, read ten lines against the image, and count what you had to fix. If it is a word or two, proceed. If it is a word or two per line, correcting will take longer than you expect, and a better copy is worth asking for.

A searchable fax is useful even when the text is imperfect, as long as you remember what a failed search means. If you search a 200-page faxed chart for "metoprolol" and get no hits, the word may still be there, recognized as "rnetoprolol". Search for a few variants and for a fragment such as "toprol" before you conclude anything. The broader checklist for bad input is in the post on triaging a low quality PDF before OCR.

Schema Law's free plan reads one document a day of up to three pages and exports the same formats as the paid plans. That is enough to run the ten-line test on a real fax from the file you are working on before you commit a whole set. The pricing page lists the limits above that.

Fax headers, cover sheets and transmission lines

Fax headers, cover sheets and transmission lines are text the machines added, and OCR reads them as if they were part of the document. They show up as a line of dates and numbers above every page of your export, and you have to decide what to keep.

The header line. The sending machine prints a strip across the top edge of each page, usually with the date and time, the sender's name or number, and a page count such as "P. 03/11". It is set in a tiny dot-built font, so it converts badly. In a Word export it becomes the first line of each page's text. In a spreadsheet of extracted text it becomes a row.

Do not delete it without thinking. That strip is often the only evidence on the page of when a document was sent and by whom, which can matter for notice, timeliness or a records custodian's declaration. Read the date and time from the image by eye, record them in your notes, and then remove the garbled line from the working copy if it is in the way. Whether a fax header proves anything in your matter is a question for the rules of your court.

The cover sheet. Cover sheets mix a few useful facts (sender, recipient, page count, a reference number) with boilerplate: a confidentiality notice in small type, a logo, check boxes. The boilerplate converts into a dense paragraph of low-quality text that will match searches for words such as "privileged" and "confidential" on every fax in the set. If you are building a searchable collection, consider leaving cover sheets out of the converted set and keeping them only in the originals.

The page count. Compare the count in the header with the pages you hold. "P. 07/11" followed by "P. 09/11" tells you a page is missing before you spend time on the rest.

Transmission reports. A confirmation page from your own machine, with a thumbnail of page one and a result code, is a separate record. File it. It rarely needs converting.

Lines and streaks. A vertical black line from a dirty scanner can be read as a column of "l" or "|" characters, or it can cut through one letter on every line. Horizontal streaks from line noise remove a slice of one text line. Find them on the image first so you know where the output cannot be trusted.

Which fax characters to proofread first

Proofread fax characters in order of consequence: numbers first, then names, then small words that change meaning. Fax damage closes the gaps inside letters and breaks their thin strokes, so characters that differ by one small stroke swap with each other.

The swaps fax resolution causes most:

• Loops that fill in: 3, 6, 8 and 0 swap with each other, and so do e, c and o.
• Thin strokes that break: 1 and 7, 5 and S, rn and m, cl and d.
• Dots that vanish: a decimal point drops out, so "1,250.00" becomes "1,25000", and a hyphen in a date range disappears.

The post on OCR errors has the longer list of error patterns that matter in legal text.

A proofreading order that fits most legal faxes:

1. Dollar amounts, including totals and the line items that feed them. Add the items and see whether they reach the stated total.
2. Dates, including the year. A date of service of 03/08 against 08/08 is a classic fax error.
3. Account, claim, policy, case and file numbers.
4. Phone and fax numbers you intend to use.
5. Names of people, drugs, companies and streets.
6. Negations and small words: "not", "no", "nor", "or", "and".
7. Body text you will quote.

Arithmetic and structure catch what reading does not. A phone number with nine digits, a ZIP code with four, a date in month 13: each is a signal to go back to the image. For the full method, including how to proofread with two people, see the section on proofreading OCR output against the original.

Asking for a better copy instead

Ask for a better copy when the content will be quoted, filed or calculated from and the fax is not cleanly readable by eye. If you cannot read a character on the image, no software can recover it, because the detail was never sent.

Signs that the fax is not worth more effort:

• You cannot tell a 3 from an 8 at 400% zoom.
• Small print has merged into gray bars.
• A streak runs through the lines you need.
• The header shows the document was faxed on to you by someone who also received it by fax.
• Pages are missing from the count.

What to ask for, best first:

1. The native file. Many faxed documents began as a word-processed letter or a report printed from a records system. A PDF saved from that system contains real text and needs no OCR at all.
2. A scan by email or secure portal. A scan at 300 dpi in grayscale carries far more detail than any fax.
3. A resend in fine mode from the original paper. Ask the sender to feed the original, not a copy, and to switch the machine to its fine or superfine setting.
4. A mailed copy when the sender has no other option.

Make the request specific. "Please email a PDF of the itemized statement dated March 8, pages 4 to 9" gets a faster answer than "the fax was hard to read". Some senders fax by policy and will decline to email records. In that case a fine-mode resend from the original is the realistic request.

While you wait, convert what you have and use it for searching and orientation only. Note in the file which passages you could not read, so nobody later treats a guess as a fact. Once the better copy arrives, run it through the same steps, and see how Schema Law handles a file from upload to export if you want the short version of that process.