1

Is it possible for Mobile Vision API to detect Chinese, Japanese and Korean? If not, is there any way to detect these language in Android?

Rushabh Shah
  • 371
  • 1
  • 7
  • 22
Rice
  • 155
  • 3
  • 11

2 Answers2

2

Google Cloud Vision API provides support for a much larger set of languages (i.e. Chinese) from an OCR standpoint: https://cloud.google.com/vision/docs/languages

Joseph Lam
  • 5,289
  • 3
  • 14
  • 13
  • You should probably mention that switching to a service from a library will also mean adding character limitations and possible fees. – Abandoned Cart Feb 03 '20 at 14:50
0

You can detect Chinese, Japanese and Korean Tesseract but you will not get 100% result.

Tutorial Link :

http://imperialsoup.com/2016/04/29/simple-ocr-android-app-using-tesseract-tutorial/

other Language Data files :

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

Sample Links:

https://github.com/rmtheis/tess-two

https://github.com/imperialsoup/SimpleTesseractExample

Janak
  • 607
  • 5
  • 23
  • Thanks. I used this before. It takes 3s to detect English but 30s to detect Chinese and Korean. Do you know why? – Rice Jun 07 '17 at 07:22
  • I have never try for other then English, but if you will provide grayscale Image then may it will take less time. – Janak Jun 07 '17 at 07:23
  • How can I improve Korean??? Is there no way to improve that by myself? Just do you need to use the provider's file? – c-an Nov 15 '18 at 09:32
  • @ChanjungKim... For better result and fast processing now you can use google vision API and it also supports the Korean language. check this link: https://cloud.google.com/vision/docs/languages and try the API : https://cloud.google.com/vision/ – Janak Nov 15 '18 at 19:53
  • 1
    Thanks. But Google Vision is not free of charge. :( – c-an Nov 15 '18 at 20:16
  • @ChanjungKim Yes, I have checked same for you. So the best option is to go with this link with Korean language pack(trained data). https://github.com/tesseract-ocr/tessdata/blob/master/kor.traineddata or https://github.com/tesseract-ocr/tessdata/blob/master/kor_vert.traineddata, You need to take very clear, zoom and grayscale image so will process faster than previous (do some image processing by your self). – Janak Nov 15 '18 at 20:20
  • @Janak A `very clear, zoom and grayscale image` is not practical for any sort of application outside of a proof of concept. – Abandoned Cart Feb 03 '20 at 14:51
  • @AbandonedCart You are right very clear, zoom and the grayscale image is not practical for any sort of application. But you will provide some image processing functionality in which a user can zoom the image and do default grayscale. So users can get the best result. – Janak Feb 12 '20 at 20:02
  • @Janak Are you demanding I provide something with the same limitations or did you intend to demand I provide something with less limitations? I should have warned you that I didn't point out your flaws without considering the argument "but what else is there?" I use https://ocr.space/OCRAPI with full-color and somewhat blurry images and have had nothing but excellent results. The caveat is that you ship the image off to a server in real time, but it is worth it to not cripple the experience. – Abandoned Cart Feb 13 '20 at 16:26