What might I do to improve tesseract accuracy for negative numbers? For whatever reason it just won't detect it correctly, even though the picture is pixel-perfect.
I'm running tesseract 4.1, using LSTM with fast tessdata. I've assigned "tessedit_char_whitelist" to "-0123456789" and I'm using segmentation mode PageSegMode.SingleChar.
Tesseract is recognizing only the dash symbol and a newline "-\n".
