Unable to detect simple math equations using pytessract

[NavpreetDevpuri]

Similar to https://github.com/tesseract-ocr/tesseract/issues/2204 and https://github.com/tesseract-ocr/tesseract/issues/1890

Environment

• Tesseract Version: tesseract v5.0.0-alpha.20200328
• Platform: Windows 10 64-bit

Current Behavior:

I download latest traineddata files from tessdata I tried following code

img = cv2.imread(file_path)
hocr_data = pytesseract.image_to_pdf_or_hocr(img, extension='hocr', lang="eng+equ")
with open("test.html", 'w+b') as f:
     f.write(hocr_data)

Input file

Output looks something like

Expected Behavior:

It should detect those equations.

Suggested Fix:

Use LaTaX as a new language to detecting math equations. Then we can easily put those LaTax math equations into hocr file as mentioned at this link

[zdenop]

tesseract is not suitable for this king of input text.

[NavpreetDevpuri]

@zdenop Why not ? Why not improve tesseract for that kind of text detections ? this is also a part of text detection. Most of the documents include some kind of equations. So, i think detecting equations are also important as detecting english.

[zdenop]

Sure you can. Just send a patch and I will merge it.

[NavpreetDevpuri]

can you help me with this How ? May be it need some testing data for training ? And how to train it ? Sorry but i am new with tesseract. Do you have any better approach ?

[NavpreetDevpuri]

i found that Let me try !

[zdenop]

No training will not help you. There are dedicated solution for such task e.g. https://mathpix.com/

[NavpreetDevpuri]

No training will not help you. There are dedicated solution for such task e.g. https://mathpix.com/

yes, i looked at it but its not free and open source

Now i am finding more solutions i got few as far now http://www.inftyproject.org/en/software.html https://github.com/UW-COSMOS/latex-ocr https://github.com/blaisewang/img2latex-mathpix

Now i am testing those

[NavpreetDevpuri]

Leaving it for now may be i will try it later

But i think its much more better if tesseract should have LaTax as an other language option for reseach documents or any kind of documents that have math equations.

[roostinghawk]

So until now you still couldn't find solution for detecting math signal? I am struggling with this problem too...

[Breakfastisready]

Hello!

Have you found any alternatives yet @NavpreetDevpuri ?

I am also interested.

[varunsawhney8]

Anyone Found Something for this problem. Please let me know.

[gulabsagevadiya]

Anyone has found any solution for this need? I am also looking for simple math equations to LaTeX

[daianaszwimer]

same here, sad to see that there are no answers to this question :(

[giovannav]

I added this configuration line of code to the function:

myconfig = r"--psm 11 --oem 3"
text = pytesseract.image_to_string(cropped_img_loc, config=myconfig, lang='eng')

It doesn't work perfectly, but it does catch most of the most common math symbols.

[adzcai]

Any updates? I'm also looking for a tool to identify equations and replace them with LaTeX. It would be enough just to get bounding boxes; then I can crop those and feed them into a different tool.

I noticed equationdetect.cpp. Is there a convenient way to use that to extract bounding boxes?

[OskarGruberPR]

... same question :)

[Kiranism]

yo! any workaround

[Shadow-Alex]

I've long investigated this area, it's quite hard and few people are working on this, you can always screen through ICDAR papers for any luck :)

[Hadar933]

still no workaround huh?

[Shadow-Alex]

Try this: https://github.com/lukas-blecher/LaTeX-OCR.

For a full ocr version, try this : https://github.com/breezedeus/Pix2Text