EasyOCR: Ready-to-use OCR with 40 languages

https://github.com/JaidedAI/EasyOCR

by vortex_ape • 6 years ago

419 69 6 years ago

github.com

polote• 6 years ago

What would be the advantage compared to something like Tesseract ?

jonatron• 6 years ago

Tesseract isn't very accurate, especially with text in photos. It works OK for scanned documents, but that's about it.

yorwba2• 6 years ago

I didn't see any accuracy comparison in the EasyOCR repo.

albertzeyer• 6 years ago

This depends on the model you use, right? As far as I know, Tesseract supports a couple of models, and you could also use a more powerful neural network in there. And if you have trained it well, it should be fine.

rexreed• 6 years ago

Are there any OCR benchmarks?

sireat3• 6 years ago

Tesseract can be very accurate (>99%), especially when you train it for your particular data set.

This does involve creating your own labeled data.

I got this 99% accuracy by performing incremental training using latest Manheim model as a base. I added about 20k lines which is not really that much. https://github.com/tesseract-ocr/tesseract/wiki

The hard part was crowd sourcing those 20k lines :)

Tesseract might not be best for photos as you said but I did not have major problems.

Of course some documents the source is so bad that a human can't achieve 99%.

Tesseract used to be quite average before they moved onto LTSM models a few years ago.

oefrha5• 6 years ago

Looking at the Chinese example, it’s kinda funny it managed to output Traditional Chinese characters when the image contains Simplified Chinese; the SC and TC versions look pretty different (园 vs 園, 东 vs 東).

hawkice• 6 years ago

Yeah, a bit strange considering traditional and simplified make up two different languages in their list of forty.

kevin_thibedeau• 6 years ago

They're rendering Unicode without any markup for language variant.

oefrha5• 6 years ago

No, these are completely different, standalone code points, not variant forms of the same code point.

What's actually happening seems to be that the ch_tra model can recognize simplified too and output the corresponding traditional version if the character isn't in the traditional "alphabet"; it doesn't work so well in the other direction.

Example recognizing a partial screenshot of https://chinese.stackexchange.com/a/38707 (anyone can try this on Google Colab, no hardware required; remember to turn on GPU in Runtime -> Change runtime type):

  import easyocr
  import requests

  zhs_reader = easyocr.Reader(['en', 'ch_sim'])
  zht_reader = easyocr.Reader(['en', 'ch_tra'])
  image = requests.get('https://i.imgur.com/HtrpZCZ.png').content
  print('ch_sim:', ' '.join(text for _, text, _ in zhs_reader.readtext(image)))
  print('ch_tra:', ' '.join(text for _, text, _ in zht_reader.readtext(image)))

Results:

  ch_sim: One simplified character may mapping to multiple traditional ones: 皇后->皇后,後夭->后夭 豌鬟->头发,骏财->发财 As reversed, one traditional character may mapping to multiple simplified ones too: 乾燥->干燥, 乾隆->乾隆 嘹望->嘹望,嘹解->了解
  ch_tra: One simplified character may mapping to multiple traditional ones: 皇后->皇后,後天->后天 頭髮->頭發,發財->發財 As reversed, one traditional character may mapping to multiple simplified ones too: 乾燥->干燥, 乾隆->乾隆 瞭望->瞭望, 瞭解->了解

Compare to the original text:

  One simplified character may mapping to multiple traditional ones:

  - 皇后 -> 皇后，後天 -> 后天
  - 頭髮 -> 头发，發財 -> 发财

  As reversed, one traditional character may mapping to multiple simplified ones too:

  - 乾燥 -> 干燥，乾隆 -> 乾隆
  - 瞭望 -> 瞭望，瞭解 -> 了解

Of course, automatic character-to-character conversion from simplified to traditional can be wrong due to ambiguities; excellent examples from above: 头发 => 頭發 (should be 頭髮), 了解 => 了解 (should be 瞭解).

joelthelion• 6 years ago

Has anyone tried it? How good is it?

MayeulC• 6 years ago

It doesn't work that bad on a few French examples I had lying around: It's doing quite well on scanned documents, even quite dense ones. Handwriting doesn't work well at all, even for simpler cases. It managed to recognize a few words from a blackboard picture, but that's hardly usable.

However, it looks like my simple example of an old "S note" export (like a lowish resolution phone screenshot) confused it a bit:

    Reglementation -> Reglemantation
    km -> kn
    illimitée -> illiritée
    limite -> liite
    baptême -> bapteme
    etc.

Overall, it works, and it is quite easy to install and use. I'd have to compare it with tesseract, but I think it's a bit better. A lot slower, though (I only have AMD devices, no CUDA). It's underusing my CPU, and maybe leaking memory a bit, though I didn't clean up.

Take that with a grain of salt, that was a quick try, I haven't tried to tune anything.

siver_john2• 6 years ago

I've recently become interested in OCR due to using Kaku on Android for trying to get better at reading Japanese. So thanks Hacker News for showing me a new version. I'd love any comments about other resources that may be good for learning. Especially because for funsies I'd like to try and develop my own.

yorwba2• 6 years ago

For learning, you could try training yourself on datasets for handwritten character recognition: http://etlcdb.db.aist.go.jp/

siver_john2• 6 years ago

Thanks! A dataset was one of the things I was dreading searching for/building. (Maybe this was super easily searchable and I'm just a goon, again I'm early stages of passing interest).

pjc502• 6 years ago

I'm going to test this on my shopping receipts, if I can get over the Windows packaging hassle, and report back ..

dclusin6• 6 years ago

From what I can tell (without having read the research papers) it looks like this is just an easy to use package for sparse scene text extraction. It seems to do okay if the scene has sparse text but it falls down for dense text detection. The results are going to be pretty bad if you try and do a task like "extract transactions from a picture of a receipt." Here's an example of input you might get for a production app: https://www.clusin.com/walmart-receipt.jpg

Notice the faded text from the printer running out of ink and the slanted text. From limited experience each of these are thorny problems and the state of the art CV algorithms won't help you escape from having to learn how to algorithmicly pre-process images and clean them up prior to feeding them into a CV algorithm. You might be able to use Google's Cloud OCR but that charges per image, although it is pretty good. Even if you use that you've graduated to the next super difficult problem which is Natural Language Processing.

Once you have the text you need to determine if it has meaning to your application. That's basically what NLP is about. For the receipts example, how do you know you're looking at a receipt? What if its a receipt on top of a pile of other receipts? How do you extract transactions from the receipt? Does a transaction span multiple lines? How can you tell? etc etc etc.

jonathanstrange• 6 years ago

So your point is that this library is not a magic unicorn that solves all problems related to OCR and natural language processing?

nacs• 6 years ago

Try reading the post. There’s a lot more there but the gist is that this is optimized for a different set of OCR uses and not the more typical scan a book/receipt cases.

D13Fd2• 6 years ago

I'm just happy to see some advancement in open source OCR for Python. Last time I had a Python project that needed OCR, I found that the open-source options were surprisingly limited, and it required some effort to achieve consistently good results even with relatively clean inputs.

Honestly I was kind of surprised that good basic OCR isn't a totally solved issue with an ecosystem of fully open-source solutions by now.

dclusin6• 6 years ago

From my experience the algorithms & implementations seem to be pretty good but the caveat is that you the developer need to be aware of all the different approaches and when it is appropriate to apply them. There just doesn't seem to be a good general purpose library that stitches them all together and knows when to use which approach based analyzing the image.

vortex_ape2• 6 years ago

Honestly I was kind of surprised that good basic OCR isn't a totally solved issue with an ecosystem of fully open-source solutions by now.

Yes! Can anyone comment on why this is the case, since OCR is proclaimed to be a solved problem?

I've always wondered why Google Lens works "out of the box" and shows great accuracy on extracting text from images taken using a phone camera, but open-source OCR software (Tesseract, Ocropy etc.) needs a lot of tweaking to extract text from standard documents with standard fonts, even after heavily pre-processing the images.

PS: Has Google released any paper on Google Lens?

driverdan• 6 years ago

Why does it have to be Python based? You can call out to other processes or services. Tesseract[1], for example, is pretty easy to work with.

1: https://github.com/tesseract-ocr/tesseract

WalterBright• 6 years ago

isn't a totally solved issue

I'm surprised, too. After all, if you can train an AI to recognize a cat, why can't it be trained to recognize a letter?

Mine, for example, works well on clean laser-printed text. It fails on anything written with a typewriter, though. (My definition of "failure" is it's quicker to retype it from scratch than fix the OCR's errors.)

I'd also love to have one that worked on cursive handwriting.

sinuhe69• 6 years ago

Yes, you're right. I tried with some scanned pages from a Vietnamese book but the result was very bad (say <5% accurate). The scans was pretty OK, though. Probably the model was not trained much for the Vietnamese language but I think it's more likely that it does not do the necessary per-processing steps.

tasogare• 6 years ago

I had very bad results on Vietnamese using Tesseract and their trained model. French output was mostly fine. I guess less attention is given to some language, and the huge number of diacritics used in Vietnamese make it harder to process too.

impostervt2• 6 years ago

I've been very impressed with the OCR on an app called Fetch, which you use to scan your grocery receipts and get points you can use to redeem for gift cards. Even if I pull a receipt out of my pocket and it's wrinkly, it still seems to read it very well.

pjc502• 6 years ago

Can you get the data from them yourself, or is it purely for them?

I've just tried easyocr on a receipt, and it's pretty bad. I've also just noticed that ASDA have a "mojibake" problem and print ú instead of £ on the entire receipt ...

maire• 6 years ago

About a year ago I surveyed the available OCR packages for receipts. This was for pristine scans (not the crumpled scan you have in your image). In my survey all OCRs failed except google cloud OCR! If there is another OCR that works I would love to know.

memexy• 6 years ago

I use TesseractOCR for general screenshot text extraction. Granted they're not receipts but Tesseract works well enough. What packages did you survey? Do you still have the data and code?

zurn• 6 years ago

Does this require a Nvidia GPU? Some modules seem to import PyTorch and Cuda libs: https://github.com/JaidedAI/EasyOCR/blob/master/easyocr/dete...

cl3misch• 6 years ago

It defaults to CPU if it does not detect a CUDA device.

catalogia2• 6 years ago

The CPU fallback taking on the order of tens of seconds on my modest i5-5250U for a few images of street signs I've thrown at it. Good enough for my purposes at least.

baicunko• 6 years ago

This is something that I find really interesting. Open-source OCR is lagging behing commercial applications and seeing someone trying ideas is always beneficial. Kudos!!

dharma1• 6 years ago

this is great, been waiting for a newer machine learning based library as an option to Tesseract.

a bit out of topic - but does anyone happen to know if there is an open-source, new school OCR library for music notation?

unnouinceput2• 6 years ago

Compared to ABBYY, how does this thing fare? I don't have the time right now to do this test and if anyone here did it I'd be thankful to share.

eastendguy• 6 years ago

It can not compete with cloud services from Abbyy, Google, OCR.space and others. But it runs locally and is open-source.

It works for sparse text on images, and for that specific use case it is better than Tesseract.

unnouinceput2• 6 years ago

I have Abby installed locally as well. I don't use it's cloud components. And I can set my own server exposing ABBYY's API's to rollout my own cloud server instead of theirs, if needs be.

say_it_as_it_is• 6 years ago

The README presents an unfinished example. How does one work with the result?

porker• 6 years ago

What are people using in mobile development (native iOS/native Android/Cross-platform e.g. React-Native) when you want accurate extraction from a fixed format-source?

E.g. poor-quality images of ID cards or credit cards, where the position of data is known.

dclusin6• 6 years ago

OpenCV is excellent for this.

tastyminerals• 6 years ago

not on mobile but as a service -- Abbyy is the market leader in OCR from documents.

oefrha5• 6 years ago

iOS has the Vision framework, can’t say whether it’s accurate enough for your use case.

https://developer.apple.com/documentation/vision/recognizing...

gooftop• 6 years ago

Anyone know how this (EasyOCR) compares with a service like AWS Textract?

saradhi• 6 years ago

What metrics do you want the comparison on?

Cost: AWS is not free vs Open Sourced

Time: AWS averages under 10 seconds vs 140 seconds on a standard Dell 7480 & 9 seconds on a GPU Google colab

Character Accuracy: Almost same on a high quality input. No comparison with AWS on a blurred camera photo like this https://github.com/ExtractTable/ExtractTable-py/blob/master/...

geonnave• 6 years ago

From the title I thought it was OCR implemented in 40 programming languages.

Yajirobe• 6 years ago

How do I train this on my own custom dataset/font?

barbs• 6 years ago

Has anyone made a desktop app with a really simple UI for detecting text in images? I'm thinking something that lives in the taskbar, lets you make a box around the text you want to read, and then returns it as plaintext?

In my job as a support engineer I sometimes get screenshots of complex technical configurations and end up having to type them in one character at a time, so this would be really handy.

Looks like maybe I could just create a wrapper around EasyOCR.

jackson1way• 6 years ago

I have the same problem every day. 20 years ago I used to have some tiny shareware tool for this but I lost it. Hope someone is able to recommend one!

nelcevest• 6 years ago

I was looking exactly for the same thing, and found this wonderful script:

http://askubuntu.com/a/280713/81372

I put it in a custom keyboard shortcut, so I just press it, draw an onscreen rectangle around any non-selectable text, and in a few seconds it goes to the clipboard.

EasyOCR: Ready-to-use OCR with 40 languages

Related Stories