OmniDocImage translation text extraction: method of identifying first and then translating
Text in pictures or scanned documents cannot be directly selected. This article explains the complete process of first recognition (OCR) and then translation, including determining the file type, organizing the original image, proofreading the recognition results, and checking the translation.
Key points of this article
- First determine whether the document is a scan or a text version: if the text can be selected, translate it directly; if not, identify it first
- The scanned PDF can be OCRed in the PDF editor of the desktop client, and then uploaded for translation after proofreading
- The recognition results must be proofread against the original image. Typos, line breaks and numerical errors will directly affect the quality of the translation.
- When you need to check, choose bilingual comparison. As long as you choose the finished product, choose translation only, and check page by page before delivery.
- For important content such as contracts, documents, medical treatment, etc., the translation must be reviewed by a person who is familiar with the field before use.
Many people get a screenshot, a one-page photo contract, or a scanned paper, and subconsciously want to directly translate the foreign language inside into Chinese. However, the text in the picture is not a character that can be selected. It is often difficult to directly hand it over to the translator, and sometimes you can only get an original picture. The correct sequence at this time is to first recognize the text in the picture into editable text, and then translate the text.
This article is written for readers who need to process pictures, photographed documents and scanned documents. It talks about the general process of image translation and text extraction. After reading, you will be able to judge which path to take with the materials you have, know how to complete "identify first, then translate" in OmniDoc, and what areas to check before delivery.
OCR refers to optical character recognition, which converts text in images into characters that can be read by computers. It does not translate itself, it is only responsible for "recognizing words"; translation is the next step. Do the two steps separately so that you can check whether there are any mistakes in the original text in the middle.
How to translate the text in the picture? One sentence conclusion
Let me answer in one sentence first: As long as the text cannot be directly selected, do OCR recognition first, proofread the original text and then translate it; for text documents that can be directly selected, just skip the recognition.
| your situation | Recommended practices | reason |
|---|---|---|
| Text can be selected and copied with the mouse | Upload translation directly | The text layer already exists, no need to identify it |
| Scanned PDF or photographed PDF | OCR first, proofread and then translate | The original text is an image and must be converted into text first. |
| A single screenshot or photo with less content | Verify the recognition results and then translate | Less content and low cost of checking sentence by sentence |
| Tables, formulas, or layouts are complex | Check the structure after identification and export to Word or Excel for processing if necessary | Recognition misalignment will disrupt the correspondence between rows and columns |
| Multiple documents need unified terminology | After unifying the proofreading standards, set the translation in the same style | Unify translations to reduce inconsistency |
Before you start: Confirm these points first
1. Determine whether the file is a scanned copy. Open the file and drag to select a piece of text. If it can be selected and highlighted, it means that the text layer exists and can be translated directly; if it cannot be selected, it can only select the entire image, which is a scan or picture, and it needs to be identified first. This step determines whether an extra link will be spent later.
2. The original picture is not clear. Recognition quality depends on input. When taking photos, try to face the paper directly, with even lighting, and avoid shadows and reflections. When the characters are too small or blurry, recognition can easily misidentify similar characters, such as "0" and "O", "l" and "1". If the original image is not clear, the subsequent proofreading workload will be significantly increased.
3. Be clear about what you want in the end. If you just want to understand the general idea, the translation alone is enough; if you want to check it with the original text, or if you want to give it to others for review, you should choose bilingual comparison. Deciding this choice before starting work can avoid discovering that the structure does not meet your needs after browsing.
4. Confirm the original language and target language. Both recognition and translation require knowing the language of the original text. Materials that are mixed in Chinese and English, such as Chinese tables with English abbreviations, are prone to language confusion during identification. Pay special attention to these areas when proofreading.
5. Acceptance of machine results requires manual review. Whether recognition or translation, machine output may be subject to errors. In particular, contract terms, drug descriptions, amounts, dates and numbers must be checked item by item. This is not a redundant step, but a necessary step before delivery.
Steps to first recognize and then translate scanned documents in OmniDoc
- Organized into PDF. If the material is multiple photos, first save it as a PDF in page order; if there is only one image, you can also save it as a PDF first and then go through the subsequent process. If the page order and file name are clear, it will save a lot of trouble during proofreading.
- Open the desktop client. Open this PDF in the OmniDoc desktop client. The client supports Windows, macOS and Linux and can be downloaded at download.omnidoc.top.
- Perform OCR on the scanned page. Use the OCR function in the PDF editor to recognize the scanned PDF and wait for the recognition to complete. If the file is already a text version, you can skip this step.
- Check the recognition results page by page. Check the original image to see if the text is complete and whether the paragraphs are broken correctly, focusing on numbers, names, models and formulas.
- Fix recognition errors. Edit the paragraph text directly in the PDF editor, change the incorrectly recognized words, and ensure that the original text is consistent with the original image.
- Upload translation. Upload the proofread document to omnidoc.top, choose bilingual or translation only, and select the corresponding style by material type, such as academic, legal or technical texts.
- Download and check the translation. Download the translation file and check key passages against the original text, especially headings, tables and sentences with numbers.
- Export when further modification is required. If the translation still needs revision, you can export the proofread content to Word or Excel for subsequent editing.
Check each item after completion
| Check items | How to see | What to do if there is a problem |
|---|---|---|
| Is the original text complete? | Check page by page against the original image to see if there are any missing paragraphs or lines. | Switch to a clearer original image and return to the OCR step to re-identify |
| Numbers and dates | Compare the amount, date, and serial number against the original image | Retranslate after correction in the PDF editor |
| Specialized nouns | Check whether the name, company name, and drug name are consistent over time | Supplement proofreading records, revise uniformly, and retranslate |
| Tables and rows | Check if the rows and columns are misaligned, and if the merged cells are correct | Manually modify based on the original image, and export to Excel if necessary |
| Check if the translation is omitted | Find any foreign language or blank paragraphs that are still retained in the translation | Check whether the corresponding page is successfully recognized, and redo OCR if necessary |
| Page count and page order | Compare the number of pages and page order between the translation and the original | If the page number does not match, verify whether the uploaded file is correct |
🎯 Which method suits you?
- Just want to quickly understand a screenshot: After recognition, read the translation directly, and match key sentences with the original image.
- Submit to others for review or archiving: Choose bilingual comparison, retain the correspondence between the original and translation, and verify paragraph by paragraph.
- To continue modifying to Word or table: After proofreading, export to Word or Excel, then continue editing.
- Handle multiple documents and maintain consistency: First, identify and standardize proofreading standards, then set translations in the same style.
❓ Frequently Asked Questions (FAQ)
1. What is the difference between image translation text and OCR?
OCR is responsible for recognizing text in images and converting images into text; Translation is responsible for converting the text into another language. Image translation is usually completed in two steps; you can check the original text in the middle when done separately.
2. Do photos of documents have to be done with OCR first?
Not necessarily. If the text can be selected, it means there is already a text layer, so just translate directly. Only when the selection is not correct and only the image is visible is recognition needed first.
3. What should you do if the recognized text is incorrect?
Corrected each part according to the original drawing. Recognition errors often occur in blurry, tilted, or small font areas. Retaking or scanning a clearer original image is usually more effective than hard editing.
4. Can the translation be used directly in formal settings?
Important content such as contracts, documents, medical or legal materials should be manually reviewed by those familiar with the relevant field before use. Machine translation can help understand the content but cannot replace formal manual review.
5. There are many tables in the picture—will it get confused after recognition?
When the table structure is complex, misalignment is likely. It is recommended to check rows and columns against the original image after identification, export them to Excel for manual adjustment if necessary, and then decide whether to return to the translation workflow.
📝 Summary
The text in the image should be identified first, then proofread, and finally translated. This order helps detect errors early on. Determine the file type, organize and clarify the original image, complete OCR on the desktop client, then select bilingual or only translation to reliably extract the image translation text.
