What Is Zonal OCR? Reading Set Areas of a Form
By Schema LawPublished Updated
Zonal OCR is text recognition limited to marked rectangles on a page. Someone draws a box around each place where a wanted value sits, such as a case number or a date, and names it. The software then reads only inside those boxes on every page with the same layout, and returns one value per box.
It is also called zone OCR. The Sedona Conference's Glossary of eDiscovery and Digital Information Management describes it as an add-on feature of imaging software that fills data fields by reading certain regions or zones of a document and placing the recognized text into the specified field. The method suits forms, because it turns a stack of identical sheets straight into rows of data.
What is zonal OCR?
Zonal OCR is a way of reading chosen areas of a page instead of the whole page. Each area, or zone, is tied to a field name. The output is a set of named values, such as "filing date: 03/14/2024", not a transcript of the document.
Two ideas are combined in it. Recognition is the ordinary reading of characters. Location is supplied by a person in advance: the software is told where to look and what to call what it finds there. That is the difference from full-page OCR, where the software must work out for itself where the text blocks are. The post on document OCR and how regions are found automatically covers that other approach.
Because zonal OCR relies on position alone, it knows nothing about meaning. It does not find "the date". It reads whatever ink falls inside the rectangle labelled "date".
Place the box on a pleading caption, then read back the venue, each party, and the statute printed in that box. A shifted sheet lets the box take the following line, often a citation, and the caption still looks filled in. Set the template aside until the zone text matches the ink.
How is a zonal OCR template drawn and reused?
A zonal OCR template is drawn once on a sample page and then applied to every page of the same layout. The template stores each zone's position and size, its field name and usually the kind of content to expect. Reuse is the point: one template can serve thousands of sheets.
The usual steps:
- Choose a clean sample of the form, scanned straight at the resolution you will use for the batch.
- Draw a rectangle around each value you want. Draw around the space where the answer goes, not around the printed label.
- Name each zone and set its type where the tool allows: digits only, a date, letters, a checkbox.
- Leave a margin. A zone that hugs the sample's text will clip the next page's text if it sits a little differently.
- Mark an anchor, if the tool supports one. An anchor is a fixed printed feature, such as the form's title or a corner mark, that the software finds first on each page so it can shift the zones to match.
- Test on twenty or thirty real pages, not on the sample. Adjust the zones that clip or pick up neighboring text.
- Run the batch and export the values, typically as a spreadsheet or a data file.
Setting a field type narrows the choices. If a zone is declared digits only, the software has no reason to return a letter "O" where a zero belongs. Fields filled in by hand are harder than typed ones, and the post on ICR and hand-printed fields explains the term used for reading them.
Legal documents that suit zonal OCR: fixed forms and cover sheets
Zonal OCR suits documents where the same information sits in the same place on every copy. In legal work that means official and standard forms, cover sheets and anything printed from one fixed layout. It does not suit contracts, correspondence or pleadings, whose text moves from page to page.
| Good fit | Why |
|---|---|
| Court cover sheets and form pleadings | Boxes for court, parties, case number and case type are in fixed positions |
| Standard-form filings with a registry or agency | One published layout, filled in many times |
| The face page of a standard incident or accident report | Date, location, report number and officer fields in set places |
| Intake questionnaires your own office designed | You control the layout, so it never moves |
| Fax and transmittal cover sheets from one sender | Same sheet every time |
| Certificates and record extracts from a single issuing office | Fixed fields, though the layout differs between offices |
| Poor fit | Why |
|---|---|
| Negotiated contracts and leases | The clause you want is on a different page, at a different height, in each one |
| Letters and emails | No fixed positions at all |
| Medical records from many providers | Each provider's layout would need its own template |
| Invoices from many vendors | Same problem, with a template per vendor |
| Any form with free-text boxes that overflow | Answers continue on attachments the template never sees |
A quick test: lay ten examples on top of one another and hold them to the light. If the answers line up, a template will work. If they do not, it will not.
Volume matters as well. Drawing and testing a template is setup work you do once per layout. That is worth it for five hundred copies of one form and not for five.
Where zonal OCR breaks: shifted pages and changed forms
Zonal OCR breaks when the ink is not where the template expects it. A page fed crooked, scanned at a different size or printed from a revised edition of the form puts values outside their zones. The software still reads each rectangle and returns whatever it finds there.
The common causes:
- Shift and skew. A sheet that enters the feeder a fraction of an inch off moves every value. Small zones lose the first or last character.
- Scale. A form faxed, photocopied at a reduced size or scanned at another resolution no longer matches the template's coordinates.
- Revised forms. Courts and agencies reissue forms. A new edition may add a line near the top and push everything below it down.
- Entries that stray. People write above the line, run past the end of a box or type in a larger font. Part of the answer falls outside the zone, and part of a neighbour's falls in.
- Stamps and handwriting across a zone. Extra marks are read as if they were the value.
- Wrong document. A page that is not the expected form at all gets the template applied anyway.
The serious feature of these failures is that they are quiet. A zone that lands half a line too low does not report an error. It returns the next field's text, or half a word, or nothing, in a tidy cell of a spreadsheet. A clipped case number looks like a case number.
Checks that catch it:
- Validate formats. A date zone should hold a date. A case number should match the court's pattern. Flag anything that does not.
- Watch for blanks and repeats. A run of empty values, or the same value on many rows, usually means a shifted batch.
- Sort by form edition before running, and keep a template for each.
- Spot-check against the page image for a fixed share of every batch, and for all values you will rely on.
Zonal OCR versus reading the whole page and extracting later
The alternative to zonal OCR is to read the entire page first and pick out the wanted values afterwards. Zonal OCR finds values by position. The second approach finds them by content, by looking in the full text for labels, patterns or meaning. Each suits a different kind of document.
| Zonal OCR | Whole page, then extract | |
|---|---|---|
| Setup | A template per layout | None per layout, though extraction needs its own rules or review |
| Copes with layouts that vary | No | Yes |
| What you keep | Only the zoned values | The full text of every page, plus the values |
| Typical failure | Right place, wrong ink | Value found in the wrong part of the text |
| Best for | High volumes of one fixed form | Mixed documents and anything negotiated |
Keeping the full text has a practical advantage in legal work. If a question comes up later about something nobody thought to zone, the answer is already in the file and searchable. With a template, unzoned text was never read.
Reading and extracting are separate steps even when one product does both, and the post on contract data explains OCR versus extraction as two different steps.
Schema Law reads whole pages. You upload a PDF, PNG or JPEG, it reads the pages into text and tables, and you export the result. Nothing in that sequence asks you to draw a template, and picking values out of the export is your own step. The export formats are the same on every plan: plain text and Word for reading, and Excel, CSV and JSON for work that needs rows and fields.
Whichever method you use, the values that matter are checked against the page image before anyone relies on them. Zonal OCR saves typing. It does not replace the comparison.