Improve detection accuracy

[dynobo]

Most of the detection quality depends on Tesseract OCR and the input image.

But there are also options/ideas in scope of this project:

• Automatically preprocess screenshot before feeding into OCR
• Tune Tesseract OCR parameters
• Apply heuristics to correct common OCR mistakes
• Improve "Magics" in detecting & transforming the OCR results

I'm happy about any ideas or pull request on this topic!

[sojusnik]

The awesome OCRmyPDF is using this optimizations, maybe they will serve as an inspiration.