Reduce the weight of eng.traineddata using only one font

[Brais Gabín Moreira]

I'm using tesseract to recognice some screenshots. I'm building this in an Android app so ~20MB of traineddata is a lot of weight. I know the font in those screenshots.

How can I reproduce the steps to generate the eng.traineddata? I want to use the same data: text, dictionary, patterns, etc. Once I have that, I'll strip out all the "useless" fonts and add the one I want.

[Quan Nguyen]

You may consider using the old versions of eng.traineddata file, one of which is only 3MB.

https://sourceforge.net/projects/tesseract-ocr-alt/files/

[Brais Gabín Moreira]

Wow! This file works as good as the 20MB! (at least in my case)

Any way it'll be great to know the steps to generate one of those files.

[Quan Nguyen]

https://github.com/tesseract-ocr/tesseract/wiki/TrainingTesseract

Some of the training source data for English are here:

https://github.com/tesseract-ocr/langdata/tree/master/eng