EM Artifact scans incorrectly

[nguyentvan7]

[legndery]

OCR is not able to read the EM main stat value:

texts
(2) ['Sands of Eon\n', 'Elemental Mastery\n']

[frzyc]

You are right, I am not sure what we can do to make this more accurate. Perhaps instead of a b/w highpass, we apply a better filter in this section? I would imaging the semi-transparent text being harder to OCR

[legndery]

I tried few standard preprocessing steps and it seems the normal gaussian blur -> dilation -> invert -> threshold is working. We can crop the stars out to be more accurate though, with the histogram we are already doing it.

Result:

Ref taken from https://replit.com/@yuispl2/GameToText#script.js this as-is worked, corresponding article: https://dev.to/mathewthe2/using-javascript-to-preprocess-images-for-ocr-1jc

Is there any regression test we can run with you know bunch of images?

[frzyc]

No regression tests (every time we bring it up seems to be more of a hassel than just taking new screenshots from in-game anyways)

Looks like good progress. the bandpass filter we been using was sort of hacked together over a weekend, and was never iterated on.