Workflow · OCR Pipeline
From ScanTailor to OCR: The Complete Page-Preparation Workflow
ScanTailor Advanced prepares scanned pages. It does not read them. Optical character recognition, and assembling the recognized pages into a finished document, happen in separate tools afterwards. This workflow treats the pipeline as four distinct jobs — capture, page preparation, OCR, and output hygiene — so each stage does the work it is actually good at.
The boundary, stated once and clearly
The official ScanTailor project description puts scanning, OCR, and assembling multi-page documents outside its scope. ScanTailor Advanced takes raw scanned page images and returns cleaned page images: split, straightened, dewarped, cropped to the content, given uniform margins, and optionally binarized. Everything that turns those images into searchable text belongs to the next tool in the chain. Keeping that boundary in mind prevents the most common workflow mistake: expecting one program to do all four jobs, and blaming it when a step it never claimed to do goes badly.
1. Capture
A scanner or camera produces raw page images. Quality here caps everything downstream — see our scanning hardware guide and the scanning DPI guide.
2. Page preparation
ScanTailor Advanced: orientation, split pages, deskew, content selection, page layout, dewarping, output. The getting-started walkthrough covers a first project end to end.
3. OCR
An OCR engine reads the cleaned pages and produces text — usually as a text layer inside a PDF, or a plain-text or word-processor file.
4. Assembly & archive
The recognized pages become the finished searchable document, while the cleaned page images are kept as masters for re-processing later.
Why preparation changes OCR results
OCR engines work from the shapes of characters. Anything that distorts those shapes makes recognition harder: a page scanned at a slight angle gives the engine slanted baselines; a curved page near a book spine bends whole lines; dark scanner-bed edges and show-through from the other side of the paper add false strokes; uneven margins and mixed page sizes confuse column detection. None of this requires a statistic to understand — a person squinting at a crooked, shadowed page has the same problem the software does.
ScanTailor’s stages each remove one class of distortion before the OCR engine ever sees the page. Deskew (including the oblique correction added in v1.1.1) straightens text lines. Dewarping flattens the curve near the spine. Select Content cuts away scanner-bed edges and stray marks. Binarization can turn a yellowed, unevenly lit page into clean black text on white — the easiest possible input for most OCR engines. Straight, clean, high-contrast pages are simply the input OCR was designed for.
Order matters: prep first, OCR second, and never OCR the raw scans “to save time” if you plan to clean the pages anyway. Text recognized from raw pages does not carry over to the cleaned images — you would be paying for the same recognition twice and keeping the worse result.
Preparing pages with OCR in mind
Choose the output mode for your OCR route
For plain text documents, black-and-white output is usually the friendliest OCR input and produces the smallest intermediate files. Keep the threshold method conservative: an aggressive setting that eats thin strokes will remove parts of letters, and no OCR engine can recognize a character whose stem was deleted at the binarization stage. Our threshold methods guide compares the available methods (Sauvola, Wolf, Fox, Window, Grad and others).
For pages where the pictures matter — diagrams, photographs, stamps, handwriting annotations — use the mixed or color/grayscale output so the illustrations survive, and let the OCR engine work on a page that still looks like the original. If you are unsure, keep a grayscale master and binarize a copy: binarization is destructive, masters should not be.
Uniform pages help layout analysis
OCR engines also analyze page structure — columns, headings, reading order. The Page Layout stage gives every page the same size and margins, which makes that structure more predictable from page to page, especially in two-column books. Running the batch workflow with consistent settings across the whole book matters more than squeezing any single page.
Choosing an OCR route
Free: Tesseract-based tools
Tesseract is a free, open-source OCR engine (Apache-2.0), maintained as an open project and used as the recognition engine inside many other tools. Driven from the command line or through a GUI front end, it is the zero-cost route and works fully offline. It recognizes text; building the final document (for example a PDF with a text layer) is handled by the tool wrapped around it — see our searchable PDF workflow.
Commercial: dedicated OCR software
Programs such as ABBYY FineReader are built around OCR as their main job, with document-layout handling and export formats aimed at offices and archives. They are a reasonable step up when you process many books, need careful table or column handling, or want recognition and document assembly in one application.
PDF-first: Acrobat and friends
If your destination is a searchable PDF anyway, a PDF editor with built-in OCR — Adobe Acrobat is the common example — folds the OCR and assembly stages into one step. This route is covered in detail in the scans-to-searchable-PDF workflow.
Whichever route you choose, judge it on a difficult page from your own book — thin paper, small print, a two-column layout — not on the vendor’s sample. Recognition quality depends heavily on the language, the typeface, and the condition of your specific pages.
Output hygiene: keep your masters
The cleaned TIFF pages that come out of ScanTailor are the most valuable files in the whole pipeline. OCR settings improve, engines improve, and your needs change — with the masters kept, re-running recognition next year costs minutes; without them, it costs a rescan, if the book is even still available. Keep three things, clearly separated:
- Raw captures — exactly as the scanner or camera produced them, untouched.
- Cleaned masters — ScanTailor’s output pages, plus the ScanTailor project file so your stage settings can be revisited.
- Finished documents — the searchable PDF, DjVu, or ebook built from the masters. These are disposable derivatives; you can always rebuild them.
Proofread like an archivist, not like a spellchecker: spot-check pages where recognition is likeliest to stumble — numbers, names, footnotes, and any page you remember being stained or torn. And set the finished document’s title and author metadata properly; a searchable file named scan_final2.pdf helps nobody in five years.
Next step: If your destination is a searchable PDF and you want OCR and assembly handled in one application, Adobe Acrobat is the route we describe in the searchable-PDF workflow. That is an affiliate link — if you buy through it, this site may earn a commission at no extra cost to you. See our affiliate disclosure. The free Tesseract route above costs nothing and is the right starting point for most personal projects.
Frequently Asked Questions
Does ScanTailor Advanced do OCR itself?
No. Per the official project description, OCR is out of scope for ScanTailor. It outputs cleaned page images; a separate OCR engine such as Tesseract, ABBYY FineReader, or the OCR built into a PDF editor turns those images into text.
Should I run OCR before or after cleaning pages in ScanTailor?
After. OCR text is tied to the exact images it was recognized from, so recognition done on raw scans is discarded when you clean the pages. Clean first, then recognize the final pages once.
What ScanTailor output is best for OCR?
For plain text, conservative black-and-white output gives OCR engines clean, high-contrast input and small files. For pages with meaningful illustrations, use mixed or grayscale/color output so the pictures survive, and keep an un-binarized master either way — binarization cannot be undone.