Zonal OCR: read one region of a page, even on a shifted scan
The invoice number sits top right. Scanned three millimeters off, the page puts half a word in your zone. Cut the zone from the answer, not from the image. Which means: a page that shifts still gives the right value for the right field.
Read one page of your own form free and see the box of every word, the pieces a zone is cut from.
Line Three lines, each with its box and a confidence of 0.99, 0.99 and 0.99. Boxes: as returned by a pipeline.
Zonal OCR on shifted scans: read form fields by region
A template with fixed coordinates
Works while every page sits in the same place. A shifted or turned page moves the zone onto half a word or the line above.
A zone cut from word boxes
Moves with the page.
Zones from word boxes in five steps
- Send the page with
format=json: a pipeline with word boxes returns a box for every word, and the page width and height give a scale. - Write your zone as shares of page width and height.
- Keep the words whose box lies at least half inside the zone.
- Build the line from those words in reading order.
- To follow a shift, move the zone with an anchor word such as "Invoice no."
Zones and templates live with you, as shares of the page (step 2).
Read turned pages upright, then cut the zone
On a turned page the boxes lie elsewhere than your zone expects. Read it upright first with the pipeline with page orientation: count on turned pages.
Word box accuracy for zonal OCR
On six clean digital test pages with 1,411 words, 98.5 % of the words carry a box that overlaps the true word by at least half along the line.
Source: own run on six clean digital pages; the truth is the text layer of the PDF.
Zone cut from the whole page: shifted scans and zone edges
The whole page is read, then the zone is cut from the answer. A shifted scan moves nothing, and every word in the zone keeps its confidence. A word at the zone edge is kept when half of its box lies inside; set that share to suit your forms.
Try zonal OCR free on your own form
Read one page of your own form free, or open the OCR demo gallery to see where the word boxes let a zone cut.
Production use, high volume or your own server?
Private use and tests are free. For a company in production, a server of your own or a pipeline built on your pages, send the kind of page and the volume, and get a written offer, or a clear no when the fit is not there.
