I have used tesseract a bit for digitizing recipes from recipe books. When it does not give good results on a first past I have found that altering the image can help a lot. Altering the image to black and white, altering the contrast, and even enlarging the image can all improve results.
I've been a subscriber to your channel for a couple of years and I like all of your videos, but this is my favorite kind of video -- about a *useful*, open source tool or utility, not specific to Arch linux, not about gaming, not about drama in the industry. I'm not saying the other kinds are bad or that you shouldn't make them, but I like this kind the best.
@Mr. Rich B.O.B Must be that they have done some sort of optimization to detect language before ocr-ing whole page. I have also seen same mistakes in tesseract as in Google Lens.
I started using tesseract for a project to gather the text off my memes hosted on my personal szurubooru (In an attempt to be able to search for set text, so you are able to actually find stuff within the thousands of images). It has been very hit or miss, sometimes it gets text right down to the punctuation, other times it gets nothing, on low res bad contrast images were I think it has no shot it gets it, on clean images it gets nothing. Sometimes doing crazy image manipulation helps, sometimes unmodified is best. What I can say is that handwritten Latin letters are impossible for it, so manga scanlations text is just blank for it, at least with the English language setting
It also doesn't support formatted text like bold, italic etc. Honestly, I usually better off with just transcribing the image text directly by typing myself with the keyboard.
I use a keybinding to invoke a script which takes the screenshot of an area, pipe it to tesseract and copy the resulting text contents to the clipboard.
Anyone, know how could I add a second language in the same command line? I tried the next command and it doesn't work: tesseract filename.jpeg - -l ara[+spa] filename.txt
OH lol, something just hit me, Google lens uses their own Tesseract OCR for extracting text and send it to your PC where you are logged in with your google account.
Excellent. I made the mistake of writing a couple thousand small notes in the stock Samsung notepad on my phone, and it turns out the garbage developers only allow you to bulk export them as PDFs instead of plain text. This will come in handy.
They're typed notes, not handwritten. Surely there's a plain text representation stored somewhere, but the program doesn't allow you to export it and they're not accessably located as text files anywhere I can find. I'm not sure how the pdfs are constructed but I'll check out pdf to text conversion too and see if that works.
This seems fairly nice for searching tango. Maybe you should also check if it can do well checking the words 1 by 1, perhaps with some other ways of framing them. I am curious how it works on Middle Eastern languages like arabic and hindi though.
I have used tesseract a bit for digitizing recipes from recipe books. When it does not give good results on a first past I have found that altering the image can help a lot. Altering the image to black and white, altering the contrast, and even enlarging the image can all improve results.
Sanitizing the image in anyway possible will always improve the results with anything like this
I've been a subscriber to your channel for a couple of years and I like all of your videos, but this is my favorite kind of video -- about a *useful*, open source tool or utility, not specific to Arch linux, not about gaming, not about drama in the industry. I'm not saying the other kinds are bad or that you shouldn't make them, but I like this kind the best.
I have many interests but this is one of them
So if Google is maintaining this project, is Google Lens just a front-end for Tesseract?
is it 😯
@Mr. Rich B.O.B Must be that they have done some sort of optimization to detect language before ocr-ing whole page. I have also seen same mistakes in tesseract as in Google Lens.
i use tesseract in python to read text and train the machine. good job explaining this 👏
A little script can be done alongside a screenshot utility to get OCR from screenshots directly to the clipboard
Haha, I did something similar with pytesseract back when I used windows. I didn't find it useful a lot though
you have video about every topic. that's awesome.
I started using tesseract for a project to gather the text off my memes hosted on my personal szurubooru (In an attempt to be able to search for set text, so you are able to actually find stuff within the thousands of images).
It has been very hit or miss, sometimes it gets text right down to the punctuation, other times it gets nothing, on low res bad contrast images were I think it has no shot it gets it, on clean images it gets nothing. Sometimes doing crazy image manipulation helps, sometimes unmodified is best.
What I can say is that handwritten Latin letters are impossible for it, so manga scanlations text is just blank for it, at least with the English language setting
It also doesn't support formatted text like bold, italic etc. Honestly, I usually better off with just transcribing the image text directly by typing myself with the keyboard.
I use a keybinding to invoke a script which takes the screenshot of an area, pipe it to tesseract and copy the resulting text contents to the clipboard.
the sad thing about pytesseract is it works as long as the background of your image is of semi-color, other than that it would mess up everything.
Anyone, know how could I add a second language in the same command line? I tried the next command and it doesn't work: tesseract filename.jpeg - -l ara[+spa] filename.txt
Thanks for the helpful video Brodie. Do you know if you can use Tesseract to convert a non-OCR'ed PDF into a PDF that contains OCR'ed text?
Have you watched Gabriel Dropout?
I have it's great
OH lol, something just hit me, Google lens uses their own Tesseract OCR for extracting text and send it to your PC where you are logged in with your google account.
Excellent. I made the mistake of writing a couple thousand small notes in the stock Samsung notepad on my phone, and it turns out the garbage developers only allow you to bulk export them as PDFs instead of plain text. This will come in handy.
Pdftotext will work better if there’s a text layer. If not… that’s nuts!
Are they saved as handwritten notes in the pdf or convert into a font?
They're typed notes, not handwritten. Surely there's a plain text representation stored somewhere, but the program doesn't allow you to export it and they're not accessably located as text files anywhere I can find.
I'm not sure how the pdfs are constructed but I'll check out pdf to text conversion too and see if that works.
This seems fairly nice for searching tango. Maybe you should also check if it can do well checking the words 1 by 1, perhaps with some other ways of framing them. I am curious how it works on Middle Eastern languages like arabic and hindi though.
Are you able to input a URL instead of a local file on your PC? This would be very useful.
OK, I just checked, and it does work with URLs. Awesome.
Any one know how to set xsane to use tesseract?
😔 still have to use google lens...
Google can't just repeat that, google drive to google docs conversion beats tesseract
Reupload?
Maybe a similar thumbnail
@@BrodieRobertson I just remember first learning about tesseract from some video, and was pretty sure it was yours
@@mskiptr maybe I mentioned it but I know I never did a video on it
See your odyssey tips and tricks for my comment
Good video.
argh, google, no, thanks
I'm a huge japan-fan, love the culture, the food and people... but anime/weebs? Cringe.
I found use --oem=1 helpful, it forces to use the new ml model which helps a lot of cases
Take a look at ocrmypdf
See your odyssey tips and tricks for my comment