How to get the best from OCR and text features

This is for version 2.0.x and 2.1.x

to be written ;-)

Just some notes for measures when default doesn't provide good results (@RaiMan needs to add some additional explanations :-)):

• Try legacy engine (OEM 0) together with tessdata from here
• Use tessdata_best models instead of tessdata_fast (causes massive performance degradation). Or just use tessdata which use integerized (faster) versions of tessdata_best without sacrificing too much accuracy.
• Use an appropriate PSM. Especially 7/8/9 if you want to detect a single line/word/character. Use convenience methods TextRecognizer.asLine(), TextRecognizer.asWord() or TextRecognizer.asChar() to get preconfigured TextRecognizerinstances for this use cases.
• If text is very small (<= 8 px) or very big (> 30px) give the OCR engine a hint using setFontSize() or setUppercaseXHeight()