OmniDocHow to choose tools for formula recognition and extraction
Formula recognition and extraction usually include screenshot recognition, whole document OCR, general chat robot and other methods. This article compares their applicable scenarios by file type and output requirements, and provides steps and verification methods for extracting formulas in OmniDoc.
Key points of this article
- When there is only a screenshot of a single formula, identification tools are faster; for the entire paper or scanned document, OCR should be done first, and then the formulas in it should be processed.
- For formulas that require repeated editing and typesetting, give priority to methods that can output LaTeX or Word formula objects, rather than just giving pictures or plain text.
- Scanned PDF, first try to select a piece of text. If you cannot select it, it means there is no text layer, and you need to do OCR first.
- No matter which tool is used, the identified formulas must be checked one by one with the original image, especially superscripts, subscripts, fractions and Greek letters.
- OmniDoc desktop client can perform OCR on scanned PDFs and export them to Word, and then process formulas in Word or UniDoc
When writing a paper review, making courseware, or organizing lecture notes, you often have to move formulas from books or PDFs into your own documents. Manually retyping dozens of formulas would take an afternoon, and it is easy to copy the wrong symbols. So many people began to search for "software for formula recognition and extraction", hoping to turn formulas into content that can be edited and reused.
This type of request seems to have only one action, but it may actually correspond to three situations: you just want to get the writing method of a formula, you want to move the formulas in the entire document together, or you want to convert the formulas in the scanned document into an editable format. The right tools for different situations vary widely. This article compares several common practices based on file type and output requirements, and explains how to extract formulas and check results in OmniDoc.
Which method should be used for formula identification? One sentence conclusion
If you only have screenshots of a small number of formulas, you can use recognition tools; if you want to extract a large number of formulas from the entire document, first confirm whether the text can be selected in the file, and then decide to use OCR the entire document or recognize them one by one. The conclusion is simple: look at the materials first, and then look at the output requirements.
| your situation | Recommended practices | reason |
|---|---|---|
| I only have one or two screenshots of formulas in my hand | Formula screenshot recognition websites or tools | Quick operation, directly get LaTeX or text writing |
| Digital version of PDF, the formula needs to be entered into Word to continue editing. | Document export or PDF editor export to Word | The text layer can be retained, and the formula can be modified in Word after exporting. |
| Scanned PDF, photo of paper book | Do OCR first, then process the formula | There is no text layer. Direct export usually does not get usable content. |
| Papers or handouts that need long-term maintenance | Unified output process of LaTeX | Formulas are saved in source code form, making subsequent typesetting and referencing more stable. |
| Just want to understand the meaning of a certain formula | Universal chatbot, with formula screenshots attached | Suitable for explanation and derivation, but the output writing still needs to be checked against the original form. |
What are the different methods?
Formula screenshot recognition websites or tools
The input to such tools is usually a picture containing formulas, and the output is LaTeX, MathML, or plain text. Its advantage is that it is highly targeted, and you can get the code for a formula in just a few seconds. The shortcomings are also obvious: it generally only processes the area you select, and will not understand the paragraphs, tables and numbers of the entire page. If you need to process dozens of formulas, taking screenshots one by one will be time-consuming, and the edges of the screenshots will easily cut off subscripts or root signs.
OCR and export the entire document
This method first recognizes the entire PDF or scan as a document with a text layer, and then exports it to Word or other editable formats. It is suitable for entire materials such as papers, textbooks, and reports. Formulas and text can be processed together. The shortcoming is that the recognition quality of complex formulas by OCR is unstable. Multi-line formulas, matrices, and formulas with many superscripts and subscripts are prone to errors. After exporting, they often need to be checked one by one and manually corrected.
Universal Chatbot
It is also the choice of many people to hand over screenshots of formulas to a general chatbot and ask it to convert them into LaTeX. It's good at explaining what formulas mean and can help you check the derivation steps. However, it is not responsible for batch processing of files, and the output results may also have symbol errors, and it cannot replace the verification of the original text. It is more appropriate to use it as a tool to assist understanding, and do not treat its output directly as the final draft.
Built-in formula function of document tool
Some document editing environments natively support the insertion of formula objects and the ability to import or export LaTeX. In this type of environment, formulas are saved as structured objects, making modification, formatting, and exporting more convenient. The premise is that the previous identification step has obtained more accurate formula content, so it is usually the step after identification, rather than replacing the identification itself.
📊 Core comparison
| practice | Input type | Output Form | Can you handle the entire document? | Check the costs |
|---|---|---|---|---|
| Formula screenshot recognition | Single formula image | LaTeX or Text Writing | No, it must be handled one by one | Low, and a single formula is easy to compare with |
| The entire OCR is exported | PDF or scanned copy | Editable documents such as Word | Yes | From medium to high, you need to check the formula section by segment |
| General-purpose chatbot | Image or text description | Text and LaTeX writing | Not suitable for mass production | In the middle, you need to compare each item with the original expression |
| Documents have built-in formula features | Existing formula content | Structured formula objects | Yes, but it depends on the previous recognition results | From low to medium, mainly check the layout |
🎯 Which method suits you?
If you only want a way to write a formula
Directly use the formula screenshot to identify it, and when selecting boxes, include the entire formula without cutting out the subscripts. After receiving the results, compare them with the original images before use.
If you want to pull out a large number of formulas from digital PDFs
First, try selecting a section of the main text in the PDF. Selecting it means there is a text layer. You can export Word from the OmniDoc desktop client, then check the formula objects one by one in Word.
If you have scanned or photographed pages of the book
Do OCR first, then process the formulas. After OCR, page-by-page verification is still required, especially for the qualifiers, subscripts, and Greek letters.
If you plan to maintain papers or handouts long-term
Prioritize processes that can store LaTeX source code. Formulas exist in source code, making future changes, layouts, and template changes more controllable.
If you mainly want to understand the formula
Give the formula screenshot to the general chatbot and ask it to explain the meaning and derivation process of each symbol. The conclusion of the explanation should be compared with the textbook or the original text; it cannot be used based solely on its answers.
How to do it in OmniDoc
- Open the OmniDoc desktop client (available for download on download.omnidoc.top) and import the PDF where you want to extract the formula.
- If the file is a scan, first perform OCR on it to make the text and formulas on the page selectable.
- Check the recognized formulas, especially subscripts, superscripts, rational expressions, and multi-line formulas. If errors are found, directly modify the paragraph text in the PDF editor.
- Export the document in Word format, check whether the formulas are complete after export, and whether the main text and table positions are normal.
- If you need LaTeX source code, you can process formulas in the UniDoc web editor (app.unidoc.top), which supports formula editing and conversion between Word and LaTeX.
- Compare each formula with the original text, confirm accuracy, and then use it in official documents.
Both OCR and formula recognition can be errors, so be sure to check each important formula before publishing or submitting.
❓ Frequently Asked Questions (FAQ)
1. Can the results of formula recognition be used directly?
It is not recommended to use them directly. Recognition results may be mistaken in subscripts, subscripts, Greek letters, and parentheses, especially for complex expressions in essays and exam materials. Comparing each item with the original text is a relatively prudent approach.
2. How to tell if a PDF is a scan?
Try using the mouse to select a passage of text in the reader. If it can select and copy text, it usually means it has a text layer; If you can't select it, it's usually a scanned image and you need to do OCR first.
3. How to save extracted formulas as LaTeX?
You can first get the LaTeX formula and then put it into a LaTeX document or an editing environment that supports LaTeX. The UniDoc web editor supports conversion between Word and LaTeX, making it suitable for this step.
4. What should you pay attention to with free formula recognition tools?
Pay attention to file size limits, whether the output contains watermarks, whether uploaded files will be saved, and whether the formula output format meets your needs. These conditions determine whether it suits you even more than the word "free."
5. Is there any way to fix the formula recognition in the scanned sample?
Try to improve the clarity of the original image to ensure the page is neat and evenly lit before performing OCR. After recognition, complex formulas are corrected one by one, saving time compared to repeatedly redoing entire documents.
📝 Summary
There is no universal approach to formula recognition and extraction: individual formulas are identified by screenshots, entire materials are exported using OCR processing, and long-term maintenance is centered on LaTeX. Before choosing, first check whether the material has a text layer, then decide on the output format. No matter which tool you use, checking formulas one by one is an indispensable step.
