c# libraries of tesseract are faster than native c++

[Iain Downs]

I mainly develop under windows and have used the emgu cv port and one of the tesseract c# ports (Charles Weld).  Both lack some features so I've build C++ libraries to better support pageiterators (and with the especial hope that bold and italic in v5 will someday work).

I'm finding that the C++ implementations are 3 or 4 times slower than the equivalent code written in c#.  The same is true if I have a C++ application and if I use static libraries or dlls for Tesseract.

I have imported Tesseract and Leptonica using VCPKG.

My code in all test cases resolves to the below or the c# equivalent:-

tesseract::TessBaseAPI* tessApi = new tesseract::TessBaseAPI();
retval = tessApi->Init(NULL, "eng");
tessApi->SetPageSegMode(tesseract::PSM_SINGLE_BLOCK);
Pix* image = pixRead(argv[1]);
tessApi->SetImage(image);
tessApi->Recognize(0);
char* text = tessApi->GetUTF8Text();

tessApi->End();
delete tessApi;

I have no idea where to look here and would appreciate any help.

Iain

[Zdenko Podobny]

Are you sure you compare apples to apples? e.g. Is not your c++ code built as debug  and C# with release optimisations?

Reasoning: Charles Weld (.NET wrapper[1]) did not port tesseract to C# but it wraps (uses) tesseract C-API in C# library. C-API  wraps tesseract C++ API. Each wrapping should cause small (maybe not even measurable) performance, but definitely wrapper could not be faster than the original library.

[1] https://github.com/charlesw/tesseract

Zdenko

so 16. 11. 2024 o 9:30 Iain Downs <ia...@idcl.co.uk> napísal(a):

--
You received this message because you are subscribed to the Google Groups "tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-oc...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/tesseract-ocr/8dbfdc4c-fa3a-4a92-9ded-4aec27384e7dn%40googlegroups.com.

[Iain Downs]

Zdenko.  Thank you.  that seems to be the explanation.  My test (C++) has gone down from 5,700 ms to 900ms.  I had assumed that the nuget package would reflect debug or release as the main project would.  It would seem that I'm wrong on that one.

Iain