Will Tesseract react well to the following training method?
- Train tesseract on a particular font and analyze its accuracy.
- Gather the letters that it does the worst on and train specifically on those in addition to regular training. For instance, if Tesseract has trouble differentiating "p" and "P", maybe train more with those by mixing and matching those characters.
- Do this for several generations.
I already pre-process the image. So right now, Tesseract is having issues with specific characters which no amount of pre-processing can solve. I am also using a Tesseract on a heavily pixelated font.
I know that Tesseract uses an "adaptive classifier". But I am unfamiliar with it.