A stamp looks like a smudge, one digit in a passport number is hard to make out, and the page leans slightly in the photo. The translation quote comes back higher than expected, even though the document is short. The reason is often that the translator isn’t being asked to translate alone: the scan may need to be read, repaired, checked, and laid out before the translation can be completed.
A blurry document scan costs more when poor image quality adds work that a clear, complete file wouldn’t need. OCR, or optical character recognition, is software that tries to turn text in an image into editable text. When the image is hard to read, the output can contain wrong words, missing details, or text in the wrong order. A person then has to compare that output with the page instead of simply translating a reliable source.
That doesn’t mean every imperfect scan automatically carries a surcharge. The added cost depends on what the provider needs to do and how their quote is calculated. Understanding the steps makes it easier to spot which part of the work is creating the extra charge and whether a better scan could reduce it.
What a scan has to go through before translation¶
A scan is an image of a page, not a document full of editable words. A translator can read the image directly, but extracting and counting the text, checking numbers, or rebuilding the page can take extra effort. OCR is one way to make the text usable; it isn’t a magic button that turns every photo into a clean, correct file.
The recognition system has to work out which marks belong to letters and words, where each line begins, and how blocks of text relate to one another. A page with a pale background, tiny print, handwritten notes, or a skewed camera angle can make those decisions harder. Microsoft lists scan quality, resolution, contrast, lighting, rotation, page size, text color, and text density among factors that affect OCR accuracy, according to its documentation checked on 2026-08-22 (Microsoft Learn).
After recognition, the extracted text still needs a check. OCR errors aren’t limited to misspelled words. Microsoft describes three kinds: a word can be substituted, omitted, or inserted even though it doesn’t appear in the source. Those errors matter in a translation because a sentence can sound perfectly natural while carrying a wrong name, date, amount, or condition. The translator has to compare the recognized text with the image, not only proofread the translated sentence.
The page layout can add another layer. A form might place a name beside a label, a certificate might have text arranged around a seal, and a report might use tables or columns. OCR can identify words while still getting their relationships or reading order wrong. Google Document AI structures recognized content into blocks, paragraphs, lines, and words, with character-level information available optionally, according to documentation checked on 2026-08-22 (Google Cloud Document AI). Reconstructing where a value belongs may take work even when the words themselves are legible.
Tesseract’s documentation says: “The quality of Tesseract’s line segmentation reduces significantly if a page is too skewed, which severely impacts the quality of the OCR.” (Tesseract documentation)
In plain English, a tilted page can make the system group the words into the wrong lines. A human may have to straighten the image, inspect the output, or read the affected area manually before translation can safely continue.
A clear scan avoids some of that repair, but it doesn’t remove every check. Translators still need to verify that the source text is complete and that the translation preserves names, numbers, and meaning. The point isn’t that software replaces those checks; it’s that a cleaner image gives the software and the person a better starting point.
Why a small image problem can create a large checking job¶
One unclear character can change the meaning of an entire field. A smudged digit in a document number may be difficult to distinguish from another digit. A faint decimal point can affect an amount. A surname with a damaged letter may be impossible to confirm from the OCR output alone. Even if the rest of the page is readable, the uncertain detail can require a deliberate comparison with the original.
A useful way to understand the extra work is to follow the file from image to finished translation:
- Inspect the scan. Someone checks whether the page is complete, readable, straight, and clear enough to work with.
- Recognize the text. OCR converts the image into editable text, if that method suits the file.
- Repair the recognition. Unclear words, missed fragments, and incorrectly grouped lines are compared with the image.
- Confirm details. Names, dates, numbers, stamps, and other important content need a closer look.
- Translate and check. The corrected source text is translated, then the translation is checked against the source.
- Rebuild the layout if needed. Tables, fields, headings, and page elements may need to be placed in a readable order.
The list describes possible tasks, not a fixed chargeable package. A provider might include OCR and its correction in a single price, charge separately for editing, or quote for layout work based on the document. Ask what the quoted amount covers rather than assuming that every provider uses the same pricing model.
A scan with one faint stamp can be more troublesome than a longer page of sharp, regular text. The stamp may contain small lettering that can’t be reconstructed from the surrounding context. A page of clearly printed paragraphs, by contrast, may be easy to recognize even when it contains more words. Image quality and document structure can matter as much as the page count.
Tesseract’s documentation defines image noise as random changes in brightness or color that make text harder to read. (Tesseract documentation)
Noise can obscure the boundaries between a letter and its background. Tesseract notes that some types of noise aren’t removed during binarization, the process that separates text from the background in a black-and-white image. A system can therefore produce a plausible-looking result while still missing or changing text.
The same issue appears with page edges. Tesseract warns that large, dark scan borders can be mistaken for characters, while cropping text too tightly and leaving no margin can also cause problems. A crop that cuts off the edge of a letter or the end of a line doesn’t just make the page look untidy; it removes information the recognition system or translator needs.
Google Document AI can return an image-quality score from zero to one and list detected defects such as document cropping, glare, or cropped text, when its relevant OCR setting is enabled. That feature belongs to Google’s product and isn’t a general standard shared by every OCR tool, according to documentation checked on 2026-08-22 (Google Cloud Document AI). The practical lesson is broader: glare, missing page edges, and cut-off text are detectable problems, not cosmetic details.
The effort grows when several defects overlap. A page can be skewed, low in contrast, and compressed at the same time. Fixing one issue may not make the text readable if another still hides part of a word. In that situation, a translator or editor may have to work through the page line by line and mark content that can’t be confirmed.
What changes the price of a scanned-document translation¶
A provider’s quote can reflect more than the act of translating words. A published price list from one Ukrainian bureau, checked on 2026-08-22, describes a model in which the base calculation depends on character count and language, while urgency, layout complexity, and editing may be considered separately. That price list is a commercial example, not a universal tariff or an industry-wide rule.
The most useful question is what extra work the scan actually creates. A quote may change because the text needs manual transcription, the OCR output needs correction, or the layout needs rebuilding. Those tasks are related, but they’re not interchangeable: correcting a misread word is different from manually entering a missing line, and both differ from placing a table into a translated document.
A price list checked on 2026-08-22 gives one dated example of how a provider can calculate a standard page and a base translation rate. It defines a standard translation page as 1,800 characters including spaces and lists a base English rate of UAH 210 per such page. Those figures describe one bureau’s published terms at that time; they don’t establish an average or current market price ([anonymous published price list; source supplied in the research dossier, checked 2026-08-22]).
Layout can also be priced with a separate multiplier. In that same commercial example, the published coefficients were 1.5 for mostly continuous text with few additional elements, 2.0 for text with tables, formulas, or illustrations taking up to 40% of a page, and 2.5 when those elements occupied more of the page. The coefficients are an example of one provider’s calculation method, not a standard surcharge for a blurry scan. A clean file with a complex table can still need layout work, while a simple page with poor image quality can need intensive checking for a different reason.
Editing can be separate too. The same price list, checked on 2026-08-22, listed editing or proofreading at 50% of the translation cost and native-speaker editing from 50%, depending on language and field. Those are dated commercial examples, not a required fee or a universal rate. A quote should make clear whether correction of OCR output is included in the translation price or treated as a separate editing service.
| Work involved | What the person may need to do | What to ask about the quote |
|---|---|---|
| OCR correction | Compare recognized text with the page and repair misread words | Is checking the OCR output included? |
| Manual transcription | Type content that OCR can’t read reliably | Are unreadable fragments priced separately? |
| Layout reconstruction | Rebuild tables, columns, fields, or headings | Does the quote include formatting? |
| Translation review | Check the translation against the corrected source | Is proofreading included or separate? |
| Urgency | Complete the work within a requested timeframe | Is there a separate urgency charge? |
That table is a checklist for understanding a quote, not a rate card. The provider needs to see the file to identify the actual work. A small but illegible handwritten note may take longer to resolve than a large block of clean print; a table may affect layout costs without increasing the effort needed to read the words.
Ask for a breakdown when the reason for an extra charge isn’t clear. Useful questions include: Is the file being charged by characters, pages, or both? Does the price include OCR correction? Will unreadable text be typed manually? Does the quote include reproducing tables and other page elements? Will the provider flag a word that can’t be confirmed instead of guessing?
Clear answers help distinguish an image-quality problem from a pricing rule. For example, a provider might charge for editing whether or not the scan is blurry. Another might include basic OCR correction but quote separately for manual transcription. Neither arrangement can be assumed from the word “translation” alone.
For more on the parts that can affect a document quote, see how translation costs are broken down. A quote based on a readable scan also tends to be easier to compare with another quote than one based on a page count alone.
How to prepare a scan before sending it¶
The best first step is to check whether you can provide a better original. A fresh scan from the paper document may solve a focus, angle, glare, or compression problem that software can’t reliably repair later. If you only have a photo or an old file, you can still improve the image before submitting it.
Google Document AI recommends a minimum scan resolution of 200 dpi and says 300 dpi or higher usually gives the best results for that OCR system, according to documentation checked on 2026-08-22 (Google Cloud Document AI). Tesseract’s documentation recommends images of at least 300 dpi for its engine, also checked on 2026-08-22 (Tesseract documentation). Those recommendations describe particular systems, not a universal legal requirement or a promise that every page at that resolution will be recognized correctly.
Resolution alone doesn’t settle the question. A high-resolution image can still be hard to read if the page is tilted, out of focus, washed out, covered by glare, or compressed so heavily that letter edges disappear. A clean lower-resolution image can be more useful than a large file with smeared text, but the right threshold depends on the document and OCR system.
Try this preparation checklist:
- Capture the whole page. Keep all text, stamps, signatures, and page edges in frame. Don’t crop so close that letters or lines touch the edge.
- Keep the camera parallel to the page. A straight, overhead image avoids the perspective distortion caused by photographing at an angle.
- Use even light. Move the page or light source to remove glare and dark shadows, especially over stamps and small print.
- Check focus at normal reading size. Zoom in on names, numbers, and the smallest text. If the letters blur together on screen, OCR may struggle too.
- Avoid unnecessary compression. JPEG is a lossy format, which means saving or compressing the image can discard image detail. Google warns that reduced file size in lossy formats can lower image quality and Document AI accuracy, according to documentation checked on 2026-08-22 (Google Cloud Document AI).
- Straighten the page. Make the printed lines horizontal rather than trying to correct a strong tilt after recognition.
- Keep a margin. Include a little space around the text instead of cutting tight to its edges.
- Check every page. A good first page doesn’t tell you whether a later page is missing, rotated, or blurred.
A practical test is to open the file and inspect the hardest area rather than the easiest one. Look at the smallest type, the faintest stamp, and any line close to an edge. If those details are clear, the rest of the page is more likely to be usable. If a name or number remains uncertain, tell the translator which part to inspect rather than silently guessing what it says.
If the file is already a scan, avoid repeatedly resizing or exporting it through apps that may reduce image quality. Keep the original image available. Sending both a compressed version and the original can help a provider compare them, but ask first which file format they can accept.
The workflow for a bad scan is explained in more detail in what to fix first in a poor-quality scan. A better image can reduce avoidable work, but it can’t restore text that was never captured in the first place.
When to rescan, and when a person needs to read the original¶
A rescan is the sensible first choice when the paper document is available and the image has a fixable capture problem. A tilted phone photo, a page darkened by a shadow, or a file blurred by compression may be replaceable with a straight, well-lit scan. Replacing the source is often safer than asking someone to infer what a letter or number might have been.
A human check against the original matters when OCR output is uncertain. Microsoft says OCR services can assign a confidence score to each recognized word from zero to one, a service-specific measure that can help identify results for human review, according to documentation checked on 2026-08-22 (Microsoft Learn). A confidence score indicates uncertainty in the recognition result; it doesn’t prove that a word is correct when the score is high or tell a translator what an unclear original actually says.
Manual reading is especially relevant for handwritten additions, signatures, faint marks, damaged pages, and text that overlaps a stamp. A recognition tool may produce no text for a fragment or return something plausible but wrong. When the original is available, a person can compare the disputed area with the paper. When the original is missing, the correct response may be to mark the text as unreadable rather than invent a word.
A scan may also be too degraded for reliable recovery. The page might be missing an edge, the print might have faded, or the image may be too pixelated for individual letters to separate. Image enhancement can adjust contrast or straighten a page, but it can’t reliably recreate information absent from the file.
Tesseract’s documentation says: “Tesseract works best on images which have a DPI of at least 300 dpi, so it may be beneficial to resize images.” (Tesseract documentation)
That recommendation is useful for the Tesseract engine, but enlarging a small image doesn’t magically add detail that wasn’t captured. Resizing may help an OCR system process an image differently; it can’t make a blurred surname certain. Ask the provider whether a new scan would help before paying for extensive manual correction.
A provider that handles scanning and recognition may include a human editing stage after OCR. A commercial service description reviewed on 2026-08-22 describes an editor correcting recognition settings and errors, with absent fragments typed manually. That description shows how extra work can arise, but it doesn’t establish that every provider uses the same workflow or charges a particular supplement.
Ask the provider to identify the uncertain parts before work begins. A short note such as “the bottom stamp is faint” or “the last digit in the reference number may be unclear” gives the translator a specific place to check. If a critical detail can’t be read, request confirmation before the translation is finalized. A guess can look polished in the translated file while still being wrong.
Common scan problems that lead to avoidable work¶
Some image defects are easy to miss when you’re looking only at the page as a whole. A scan can seem readable at a glance while hiding an issue that becomes important during recognition or translation. Checking the likely trouble spots first can prevent an avoidable round of questions and corrections.
A tilted page. Skew changes how software detects lines. Tesseract says that a page that is too skewed can significantly reduce line segmentation quality and affect OCR, according to documentation checked on 2026-08-22 (Tesseract documentation). Rotate the page so printed lines are horizontal before sending it.
Glare over a word or seal. A reflection can wash out the ink it covers. The rest of the page may look sharp, but the missing letters still need checking. Google Document AI identifies glare as one of the image defects its quality assessment can report when the relevant setting is enabled, according to documentation checked on 2026-08-22 (Google Cloud Document AI).
A cropped edge. Cutting off a small part of a page can remove letters, page numbers, or the end of a line. Tesseract also warns that large dark borders can be interpreted as characters, while text cropped without a margin can create its own recognition problems (Tesseract documentation). Keep the whole page and a little space around the text.
Overcompressed images. A file may be small and quick to send but have smoothed or blocky text. Google warns that reducing the size of lossy formats such as JPEG can reduce image quality and the accuracy of Document AI, according to documentation checked on 2026-08-22 (Google Cloud Document AI). Keep the original scan and avoid repeatedly exporting a compressed copy.
Tables and forms treated like paragraphs. OCR might recognize a value but lose the connection between that value and its label. A form field can then be assigned to the wrong row, or a table can become a stream of words. Check relationships between labels and values, not just whether the OCR text contains the expected words.
The translation checked without checking the source. Spelling and grammar checks on a translated document won’t reveal a word that OCR omitted before translation. Microsoft describes substitution, omission, and insertion as OCR error types, so a reliable check needs to compare source text with the original image (Microsoft Learn).
A guess left unmarked. When no one can read a fragment, filling in the most likely word can create a false certainty. Ask the translator to flag text that can’t be confirmed. A clearly marked uncertainty is easier to resolve than a confident-looking error.
For documents with complicated tables, columns, or page elements, how to translate multi-column layouts without reflow chaos explains why reading order and layout need attention as well as the words. A clean scan helps, but a structured page may still need careful reconstruction.
FAQ¶
Why does a blurry document scan cost more to translate?¶
A blurry scan can need more than translation: OCR correction, manual reading, checking names and numbers, or rebuilding the page layout. Whether those tasks add a charge depends on the provider’s pricing method and what the quote includes.
How can I prepare a document scan to avoid extra translation costs?¶
Send a complete, sharp, straight, evenly lit scan, and keep a small margin around the text. Avoid glare and repeated compression, then ask the provider whether a better scan would replace the file already submitted.
What scan resolution is needed before OCR?¶
Google Document AI recommends a minimum of 200 dpi and says 300 dpi or higher usually gives the best results for its system; Tesseract recommends at least 300 dpi for its engine. The recommendations are system-specific and don’t guarantee accuracy for every page (Google Cloud Document AI; Tesseract documentation).
Can a blurry document be translated without scanning it again?¶
A person can sometimes read and verify unclear text from the existing image, especially when the original document is available for comparison. Missing, cropped, or heavily degraded text may remain impossible to confirm, so ask the provider to identify uncertain passages rather than guess.
What affects the cost of translating a scanned document?¶
Character count, language, OCR repair, manual transcription, editing, urgency, and layout work can all affect a quote, depending on the provider. Ask which tasks are included, which are separate, and how unreadable fragments will be handled.
Need these documents translated?
Leave your number and we will call back, tell you which certification your institution needs, and name the price. Translation, translator review and notary certification.