Calculation of confidence levels

[Purushotham Rao Eravalli]

Is there a way to figure out how does tesseract calculate confidence levels for characters or words. As of now if we observe we are getting higher confidence even though the text is wrong, lower confidence even though the text is correct. So it might be helpful if we know. And is there anyway to tweak the process and retrain it?

[Mobeen Ali]

Yes, use HOCR output form it contains the confidence value of each word