Hi Joe,
add this to your code:
gray = preprocess(gray_base.copy())
cv2.imshow(f"Detected unique hex codes {idx}", gray)
cv2.waitKey(0)
you'll see the problem: two preprocessing methods generate junk, the other
two are almost identical and the problem is a too high threshold: try 85
(the font on the right is lighter). Always visualize the image after EACH
preprocessing sub step to see where it starts to go bad.
For a problem like this with fixed patterns/scale/colors, etc. I would use
cv2.matchTemplate:
https://docs.opencv.org/4.x/d4/dc6/tutorial_py_template_matching.html
it should be 100% accurate.
You can also "blend" each code into a blob with dilate/erode, run a
findComponents, crop out each region, run matchTemplate on each fragment.
But I think the grid location is fixed so I would just use two nested loops
to crop the codes at exact locations and template match.
If you have a smartphone picture as input you may want to use the first
method so you can do a cv2.warpPerspective to align/rescale the grid to a
fixed size/location before proceeding.
But matchTemplate should handle multiple matches fine so there is no need
to complicate things.
Please let me know how it ends.
Bye
Lorenzo
Il giorno lun 30 giu 2025 alle ore 05:32 JoeBlack <[email protected]>
ha scritto:
> Source Code: https://pastebin.com/7XvxnZa6
>
> I'm trying to OCR a screenshot of a "breach protocol" minigame from
> Cyberpunk 2077. In the game, there's a grid of two-character hex codes
> (e.g., 1C, 55, etc.). My goal is to detect all expected hex codes from a
> known list. I'm using multiple preprocessing methods (thresholding,
> adaptive thresholding, sharpening, etc.) and morphological closing to
> improve recognition. Despite this, some codes in the main matrix are still
> missed completely.
>
> Below is my code that runs several preprocessing passes, combines unique
> tokens across passes, and draws rectangles around recognized hex codes. Any
> suggestions for improving detection accuracy, especially for codes missed
> in the main matrix, would be greatly appreciated!
>
> --
> You received this message because you are subscribed to the Google Groups
> "tesseract-ocr" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> To view this discussion visit
> https://groups.google.com/d/msgid/tesseract-ocr/070afc51-4c9f-4692-b05a-11b06a13964dn%40googlegroups.com
> <https://groups.google.com/d/msgid/tesseract-ocr/070afc51-4c9f-4692-b05a-11b06a13964dn%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
>
--
You received this message because you are subscribed to the Google Groups
"tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion visit
https://groups.google.com/d/msgid/tesseract-ocr/CAMgOLLy1rpn8NDnPSYh8HN9hbgYzoUbkU-4WOcVH1EA%2BhXVbKA%40mail.gmail.com.
