Android tesseract no text out at recognise()

[scubess]

Hi @rhardih ,

I compiled tesseract 4.0.0 for android and produce shared libraries using Building from one of your Android Docker file.

I wrote a simple JNI wrapper and using ndk-build with android.mk file, I build a shared library that goes into my application. I can see, tesseract has been initialised and can get the version 4.0.0 with no issue. I am seeing no crashes but at the same time, I can't see any text has been recognised or no error is thrown.

I followed the instructions and I am sure, I downloaded the right version of trained data file into tessdata folder. Even I tried with different versions of .traineddata, didn't make any difference.

To note, when I use tesseract 3.05, I can see the code runs with no issues. I can't understand what's going wrong here. I have been trying for days and I couldn't find the issue.

I tried to post this question in tesseract. Sadly, its been closed as they said they cant provide any help for throd party projects.

I attached the sample app with JNI wrapper please click the link for the sample android application

Environment: Tesseract
Tesseract Version: tesseract 4.0.0

Current Behavior:
No text recognised at in api->GetUTF8Text()

Expected Behavior:
returns recognised text from the image

Look forward yours help!

[rhardih]

Are you running the LSTM engine or the legacy one?

I remember something about the new LSTM engine possibly not working on Android due to being too memory demanding. I've only used the legacy engine myself.

This issue has a bit more info: rmtheis/tess-two#196

[scubess]

Hi @rhardih ,
thanks for reply. I haven't set the engine mode at all. I beleive it goes to default. I tried in the morning to set the engine mode to
apiTest->SetVariable("tessedit_ocr_engine_mode", "OEM_TESSERACT_ONLY");

It didn't made any difference.

[alexcohn]

Hi @scubess, it would be much easier to help you if you post the bitmap that is passed to OCREngineImp.recognition()

[scubess]

Hi @alexcohn ,
Thanks for helping me. Please find the attached image that i am sending for recognition and also i creating one via code in the sample app

[alexcohn]

Maybe I am missing something, but your bitmap should be 640x480, and here you post a PNG of size 1280x960 pixels. If by any chance this image is passed to tesseract with wrong size and/or format, it won't have a chance to find text there.

[scubess]

Hi @alexcohn,

Sorry, It's my fault. I didn't made it clear to you. The bitmap image I am passing is created via code in the sample app. But as you asked the image, I have another image which I created and load from assets folder. Tesseract didn't recognise both images. The one I sent to you is loading from Assets folder.

[alexcohn]

Exactly, so I thought, maybe the bitmap that is created via code is not good. With a png from assets, it's easier to debug the engine.

[scubess]

Hi @alexcohn ,
Just let you know, the info may be helpful. I have been trying to send bitmap images from Camera. The recognition works for v3.05 of Tesseract. But it's not working for tesseract 4.0.0. I put the code in the issue/test_from_file branch in the sample app.

Thanks again for looking into it.

EDITED: @alexcohn I made a change to load image from asset folder for you now

[scubess]

Hi @alexcohn @rhardih ,

When I build tesseract 4.1.0 and use 4.0.0 LSTM (tessdata_fast) models, it worked!

Thank you so much for your work & help!

[rhardih]

Hi @scubess, sorry for the late reply. Been a busy week. Good thing you got it to work. I've added an issue for bumping the default 4.x to be 4.1 here #12. Will get on it soon hopefully. 👍

[scubess]

Thank you @rhardih