OmniDoc# item-by-item troubleshooting and repair
Explain the difference between OCR recognition failure and recognition error, list the reasons for common symptoms such as empty results, garbled characters, typos, etc., and provide troubleshooting steps to first repair the source file and then re-recognize it, to help you process scanned and photographed documents into usable text.
#Causes and solutions for OCR recognition failure
Key points of this article
- Recognition failure means that no usable text is output, and recognition error means that the text is output but the characters are wrong. The reasons and correction methods of the two are different.
- Most failures are caused by original image distortion, shadows, low resolution or inconsistent language settings. Repairing the source file first is usually more effective than repeatedly adjusting settings.
- Before processing, use the method of selecting text to determine whether the page is an image, which can quickly eliminate the misjudgment of "actually there is a text layer".
- Oversized or multi-page files can be split by chapters and identified in batches, and then the results can be merged to reduce the probability of errors midway.
- OCR results may still have errors, and key content such as contract amounts, formulas, and proper nouns need to be checked item by item.
Many people submit scanned contracts, papers, or photographed handouts to OCR (Optical Character Recognition) for processing. As a result, either no word is recognized, or the screen is full of garbled characters, missing characters, typos, or even a direct prompt that the processing failed. These situations all seem to be "OCR is not easy to use", but the reasons and methods are different. This article is written for people who need to turn scanned documents or pictures into editable and searchable text. It will help you determine where the problem lies and give you a solution that you can follow.
First distinguish between two concepts. Recognition failure means that the engine did not output usable text, which may be because the result is empty, processing is interrupted or an error is reported; Recognition error means that the text has been output, but the characters are wrong, such as "0" and "O" are confused, """ becomes "about", and the table is misaligned. The former is mostly related to the file itself, while the latter is mostly related to image quality and language settings, and the troubleshooting order is also slightly different.
After reading this article, you can first find out the cause by comparing the symptoms, and then follow the steps to correct the source file and re-identify it.
OCR recognition failure solution: Repair the source file first or change the settings first? One sentence conclusion
In most cases, it is more effective to repair the source file first and then re-identify it than to repeatedly adjust the settings.
| your situation | Recommended practices | reason |
|---|---|---|
| Full page scan with skew or shadows | First correct the angle, crop the black edges, improve the contrast, and then re-OCR | Slants and shadows blur the edges of strokes, making it difficult for recognition engines to separate adjacent characters. |
| When taking pictures with a mobile phone, the text is too small or blurry | Reshoot, tile the light evenly, or scan with a higher resolution instead | When the resolution is insufficient, font details are lost, and typos and missing characters will increase significantly. |
| Chinese and English mixed or multi-language documents | Confirm the main language and process the mixed content into segments by language. | The language setting does not match the content, which can lead to widespread misrecognition. |
| The file is large and has many pages | Split by chapters, identify in batches, and finally merge | When a single file is too large, it is prone to timeout or interruption. |
| File is password or permission protected | Ask the file provider for an unencrypted version before processing | The protected file cannot read the page image normally and needs to obtain a legal readable version first. |
First determine what kind of problem it is
Looking at the symptoms before troubleshooting can save a lot of ineffective attempts.
| symptoms | Common reasons | How to confirm |
|---|---|---|
| The result is empty and there are no words | The page itself is a pure image, or the resolution is too low or the color is reversed. | Try to select a paragraph of text with the mouse in PDF, but if it is not selected, it is an image page. |
| A large number of garbled or strange symbols | The language setting is incorrect, or the original text uses a special font | See if the garbled codes are concentrated on the characters of a certain language, and compare them after changing the language. |
| Typos, missing words, line breakers | Scan skew, shadow, noise, or low resolution | Enlarge the original image so that you can clearly see a single stroke to see if the stroke is adherent, empty, or obscured by shadow. |
| Form or formula identification disorder | The layout is complex, the lines are uneven, and the formula is mixed with the text. | Only check the table or formula area to identify separately to see if the result is improved. |
| An error is reported in the middle of processing or there is no response for a long time. | The file is too large, too many pages, or the file itself is corrupted | Can you open it normally with other readers; can you open a few pages and deal with them separately? |
Operating in the following order, the easy before the difficult, usually solves most of the problems.
- confirm the file type. try to select a paragraph of text in PDF. Selected, indicating that the page already has a text layer, can be directly copied or searched, does not necessarily need to be selected by OCR;, is the image page, need to identify.
- check the clarity of the original image. Open the original image and enlarge it paragraph by paragraph to see if the strokes of the text are clear. If you can't even see it yourself, there is a good chance that OCR will go wrong, so you should change to a clearer version first.
- correct scan quality. rotate or straighten the skewed page, cut out the scan frame, fingers, desktop and other interference, remove the shadow, and improve the contrast appropriately. These operations can be done with the scanner's own functions and common image editing tools.
- adjust the resolution or re-shoot. when taking photos, the phone should be laid flat as far as possible, so that the light is uniform and the lens is facing the page, so as to avoid reflection and perspective distortion. Improve the resolution when scanning, but the document does not need to be enlarged infinitely, and the handwriting of the text shall prevail.
- confirm the language settings. choose the main language of the document; when mixing English and Chinese, check whether English proper nouns and numbers are correctly identified.
- split and try again. when there are a large number of pages or large files, divide them into small files according to chapters or several pages, identify them separately, and then merge the results. In this way, you only need to redo some of them after reporting the error.
- reprocess the revised file with OmniDoc. * after uploading the modified scan PDF in omnidoc.top, you can translate the document directly, or you can OCR the scan PDF in the PDF editor of the desktop client (download.omnidoc.top), and then compare the new and old results to confirm which step has improved the effect.
- check the key content. errors may occur in OCR and machine translation. The contract amount, date, formula, person's name and technical terms should be compared with the original item by item and should not be used directly as the final version.
# how to avoid the next occurrence
- uniform standard for scanning : fixed resolution, color mode and page orientation to avoid uneven quality of the same batch of files. - fixed position when taking photos : use a bracket or desktop to fix the phone, facing the paper, and turn off the light that will cause reflection. - read one page before processing the whole book : before batch processing, test one or two pages to make sure the language and settings are correct, and then process all pages. - keep the original document : save another copy before correcting the skew and trimming, so that you can go back to the original drawing and do it again at any time. - create a checklist : for important documents, make a list of numbers, amounts, dates and proper nouns, and check them item by item after identification.
# frequently asked questions (FAQ)
1. What's the difference between OCR recognition failure and recognition error?
Recognition failure is the absence of available text, such as empty results, error reports, or processing interruptions; recognition errors are text, but there are typos, missing words, or typesetting errors. The former is mostly related to file type, size and permissions, while the latter is mostly related to image clarity and language settings.
2. Which is easier to identify successfully with a mobile phone or a scanner?
The scanner usually produces a smoother page, more uniform light, and a more stable recognition success rate. Mobile phone photography should pay attention to flat, positive, uniform light, otherwise perspective and shadows will significantly affect the results.
3. Fixed scan quality, there are still typos, is it the problem with the tool?
Not necessarily. The handwriting adhesion of the original picture, paper wrinkles or too light ink may limit the recognition effect. You can first go back to the original image to check the clarity, and then decide whether to rescan. For important content, no matter what tool is used, it should be checked manually.
4. Can the recognized text be directly used in official documents?
It is recommended to check it before using it. OCR results may have errors, and numbers, formulas and proper nouns in contracts, financial and academic materials in particular need to be compared item by item with the original.
# Summary
The failure of OCR recognition is mostly due to the quality of the original image, page type or language setting. First find the cause according to the symptoms, and then correct the source file to re-identify, the success rate is usually higher. Fix the inspection steps and preventive practices, and later processing the scanned parts will save a lot of rework time.
