Tesseract crashes when processing certain documents

[TerryZH]

Environment

• Tesseract Version:
  Tesseract Open Source OCR Engine v4.00.00dev-692-gad5ee184 with Leptonica

• Platform:
  Platform: Linux 4.9.43-17.39.amzn1.x86_64 defect issue #1 SMP x86_64 GNU/Linux

Current Behavior:

The following command will crash in the above stated environment:
tesseract /tmp/tr_tmp.jpg /tmp/tr_tmp --tessdata-dir /var/task/tessdata --psm 12 --oem 2 -l eng hocr

tessdata is legacy data from tesseract-ocr/tessdata

Crash error: Assert failed:in file ../ccutil/unicharset.h, line 513

related jpg file:

related binaries:
Archive.zip

Expected Behavior:

When tesseract v4 using --oem 2 and legacy trained data, error message of missing LSTM data should be printed instead of crashing.

Suggested Fix:

n/a

[Shreeshrii]

The latest traineddatas (tessdata_best and Tessdata_fast) do not support legacy tesseract engine, so --oem 0 and --oem 2 are not supported.

However, program should not crash but rather give an error message.

[TerryZH]

Thanks for the update. Where can I get the 'latest traineddata' please? I got my data from https://github.com/tesseract-ocr/tessdata/

[Shreeshrii]

Please see https://github.com/tesseract-ocr/tesseract/wiki/Data-Files#updated-data-files-for-version-400-september-15-2017 ShreeDevi

…

____________________________________________________________ भजन - कीर्तन - आरती @ http://bhajans.ramparivar.com

On Mon, Oct 23, 2017 at 10:21 AM, Tao ***@***.***> wrote: Thanks for the update. Where can I get the 'latest traineddata' please? I got my data from https://github.com/tesseract-ocr/tessdata/ — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1181 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AE2_o1A6Ocre9MuS8eUo5CfbBl_Ibfk2ks5svBtUgaJpZM4QCEnk> .

[TerryZH]

Thanks for your reply. Just to be rigid. I was using tessdata, which supports oem mode 2 according to the wiki.

[amitdo]

The latest traineddatas (tessdata_best and Tessdata_fast) do not support legacy tesseract engine, so --oem 0 and --oem 2 are not supported.

Although I wrote something similar to the above remark in the wiki, since commits [1] and [2]. --oem 0 -l osd works with 'best' and 'fast'. Not sure about --oem 2.

[1] tesseract-ocr/tessdata_best@f1d1268
[2] tesseract-ocr/tessdata_fast@139ff12

[TerryZH]

Since the discussion is a bit side-tracked, I'm repeating the problem. The original� comment is also updated.

Tesseract v4.00.00dev-692-gad5ee184 crashes when using --oem 2 and tesseract-ocr/tessdata.

[PaniniGelato]

The master branch (commit ad5ee18) runs good with your image @TerryZH

However, tesseract should not crash

[TerryZH]

@PaniniGelato Are you sure? As reported in the "Environment" section, this crash happens on the build from master commit ad5ee18. And I can still reproduce the crash. Did you use the same tessdata as described in my previous comment?

[zdenop]

It does not crash on windows (the latest code):

>tesseract.exe i1181.jpg i1181 --psm 12 --oem 2 -l eng hocr
Tesseract Open Source OCR Engine v4.0.0-beta.4 with Leptonica

I will try to test it on linux later. In meantime please check if you are using the latest traineddata...

[stweil]

This is the well known assertion ASSERT_HOST(contains_unichar_id(unichar_id)); which occurs with certain images, but only when using both legacy and LSTM OCR engine.

[zdenop]

duplicate to #1205.
Anyway it should be fixed with 9efedc1.