https://bugs.kde.org/show_bug.cgi?id=478426

--- Comment #18 from Noah Davis <noaha...@gmail.com> ---
(In reply to Niklāvs Koļesņikovs from comment #17)
> If the search query asked for C++, then, indeed, the list was probably
> short, since by far most such software is written in C, CUDA or maybe even
> Python with numpy, pytorch or similar framework for SIMD or GPU
> acceleration. However by asking for hardware acceleration, it's implied that
> the actual processing would be done by some kind of a shader, so all it
> would really take was attaching a GLSL or SPIR-V shader implementing the
> desired mathematical kernel, which is something Qt can already do. Although
> probably not directly with QImage, so it would likely need to be copied to
> QOpenGLTexture or a similar type and after processing copied back to QImage.
> 
> Regarding OCR, I'm quite certain that OpenCV uses Tesseract OCR (
> https://github.com/tesseract-ocr/tesseract ) which is written in C++ and
> could be used directly. That being said, my gut feeling is that there's
> probably something better out there, just maybe not published yet, since the
> best I could quickly find was using one of the generative adversarial
> networks (GAN) for cleanup before feeding the processed image into a 
> convolution neural network (CNN) based OCR i.e. Tesseract. However GANs are
> quite amazing and I'd expect them to eventually replace CNNs for OCR
> purposes.
> 
> In short, there's nothing magical about OpenCV and the desired bits can
> either bit assembled from existing projects or directly implemented in Qt or
> KF, since there's probably more than just Spectacle that would be greatly
> improved with some graphical or compute shader based features (I'd certainly
> love either advanced scalers or ANN processing in Okular and Gwenview).

Do you know of any image processing libraries besides OpenCV that have good
documentation? If I was determined, I could figure out how to use most well
made libraries eventually, but I want to get this over with as quickly as
possible so that I can move on to other things. I'm currently not interested in
trying anything that isn't fairly mature and widely used.

Another library I looked at briefly was CImg, but their documentation website
doesn't even seem to work.

-- 
You are receiving this mail because:
You are watching all bug changes.

Reply via email to