ICR vs OCR: Handprint Recognition Explained
By Schema LawPublished Updated
The difference between ICR and OCR is the kind of writing each was built to read. OCR, optical character recognition, reads type that a machine printed. ICR, intelligent character recognition, reads letters and digits a person printed by hand, usually one per box on a form. Neither label promises that joined-up cursive will be read correctly.
What is the difference between ICR and OCR?
OCR reads type a machine put on the page: a printed lease, a typed letter, a fax of either. ICR reads characters a person wrote by hand and kept separate, such as a date of birth printed into boxes on an intake form. The input differs, and so does how much checking the output needs.
| OCR | ICR | |
|---|---|---|
| Reads | Machine print in ordinary fonts | Hand-printed capitals, lower case and digits |
| Typical page | Contract, pleading, statement, letter | Application, claim form, questionnaire |
| Where the characters sit | Anywhere on the page, in lines and paragraphs | In fields, often one character per box |
| Why it is harder or easier | Letter shapes repeat exactly | Every writer forms each letter a little differently |
| What you check afterwards | Names, numbers, faint or small print | Every field that matters, one by one |
A worked example makes the split concrete. Suppose a fictional client, Dana Whitlow, fills in a paper intake form for the fictional firm Reyes & Calder. The form's printed questions ("Date of incident", "Employer name") are machine print, and that is OCR's job. The answers she prints in block capitals inside the boxes are handprint, which is what vendors mean by ICR. The paragraph she writes in flowing script under "Describe what happened" belongs to neither label in its original sense. One page, three kinds of writing, three levels of confidence you should place in the text that comes back.
What does ICR stand for and where did the term come from?
ICR stands for intelligent character recognition. It is a trade term from forms processing, used for software that reads handwritten characters where plain OCR reads print. No single definition is shared by every source, so the word tells you less than a test page does.
Two published descriptions show how loosely the term is drawn. The Sedona Conference's Glossary of eDiscovery and Digital Information Management defines ICR as the conversion of scanned images to computer-recognizable codes by software that defines the rules for the conversion, and adds that it is "helpful for interpreting handwritten text". The same glossary lists "Handwriting Recognition Software (HRS)" as a separate entry, and its ICR and OCR entries each point to the other and to HRS.
Research data draws no hard border either. The National Institute of Standards and Technology publishes a collection of hand-filled sample forms from 3,600 writers. NIST calls it the Handprinted Forms and Characters Database and describes it as intended for "handprint document processing and OCR research". In that description, reading handprint is a branch of OCR, and the word ICR does not appear.
The practical reading is this. "ICR" on a product page tells you the vendor believes the tool handles handwriting of some kind. It does not tell you which kind, how well, or on what forms. The adjective "intelligent" names no particular method.
Hand-printed boxes on an exhibit often hold the party name and the injury date, while the printed instructions cite a statute. Check both readings against the page before you rely on search. Miss the handwritten name and the claim stays hidden in the file.
Hand-printed boxes versus cursive: what ICR was built for
ICR in its original sense was built for constrained handprint: separate characters, written inside marked fields, by someone who was asked to print clearly. Cursive is a different problem, because the letters join and the software must first decide where one ends and the next begins.
Forms are designed to make handprint easy to read. Boxes or comb marks hold one character each. Instructions say "use block capitals" and "black ink". A field labeled "ZIP code" can only contain digits, which removes most of the guesswork. When software knows where a field sits on the page, it can also read just that area, an approach covered in the post on what zonal OCR is.
Handprint on a legal form gets harder when:
- The writer ignores the boxes and lets letters touch or cross the lines.
- A field holds free text, such as a street name, where any spelling is possible.
- Lower case and capitals are mixed, so "l", "I" and "1" look alike.
- The form was faxed or photocopied several times before it reached you.
- Corrections are written over the first attempt.
Free-flowing handwriting, such as a witness statement in script or a lawyer's margin note, sits outside what the ICR label first covered. Whether software can read it at all, and how far to trust the result, is the subject of the section on whether software can read handwriting.
OMR and other marks that are neither ICR nor OCR
OMR, optical mark recognition, detects whether a mark is present at a known spot on a page. It reads no characters. A ticked checkbox, a filled bubble on a questionnaire and a struck-through option are marks, so they fall under OMR, not under ICR or OCR.
The distinction matters because the output is different in kind. Character recognition returns text. Mark recognition returns a yes or a no for each position, and it needs to know in advance where the positions are.
| What is on the page | Term | What software returns |
|---|---|---|
| Printed question text | OCR | The words |
| Block capitals in boxes | ICR | The characters, to be checked |
| Checkbox or bubble | OMR | Marked or not marked |
| Barcode or QR code | Barcode reading | The encoded value |
| Signature | None of these | At most, whether ink is present in the signature area |
| Ink stamp or seal | OCR on the lettering, if legible | Some or none of the words |
Two cautions apply to legal forms. A general-purpose OCR tool that is not doing mark recognition may turn a checked box into a stray character, such as an "X" or a "v", or drop it. And a box can be ambiguous on paper: a faint tick, a mark that touches two boxes, or one that was crossed out and replaced. Always read ticked answers against the page image. The post on form fields and checkboxes in police reports shows how that plays out on a real kind of form.
Does the ICR label still matter when you choose a tool?
The label matters less than a trial on your own forms. A product may describe one reader for print and handwriting and never use the word ICR, or it may list ICR as a separate module. Either way, the name does not tell you how the tool does on the pages in your file.
Questions that get you further than the acronym:
- Which kinds of handwriting does the vendor say the tool reads? Boxed handprint, free handprint and cursive are three different answers.
- Is handwriting read by default, or is it a setting or an add-on? If it is an add-on, ask what it costs.
- Does the tool need a template for each form? A template helps on a form you process hundreds of times and is a burden on one you see twice.
- What happens to a field it cannot read? Blank, a guess, or a marker are very different outcomes for a date of birth.
- How are checkboxes reported? Ask to see the output for a page that has them.
Then run the test. Pick five filled-in forms from a closed matter, including one with poor handwriting and one that was faxed. Convert them, and compare every handwritten field with the paper. Count the fields that came out wrong, and note which kind of writing they were.
Schema Law does not use the ICR label, and this post makes no claim about how it reads handwriting. The free plan converts one document a day of up to three pages, so you can upload a filled-in form and judge the text for yourself; the pricing page lists the limits on larger files. For the rest of the vocabulary you will meet while comparing tools, the OCR glossary for lawyers defines each term in a line and points to the post that explains it.