White text on blue, green background. different result

[L ht]

Hi, 

I cropped two buttons from a screenshot. The images contain white text on blue or green backgrounds, but I’m getting different results.
Using Ubuntu Tesseract 5.4.0 with the following commands:
tesseract detected.png - return "SIGNIN WITH FACEBOOK"
tesseract undetected.png -  return empty string

Can anyone explain why this is happening and how to detect both? Thanks.

[Zdenko Podobny]

Did you read the tesseract documentation (before asking the forum)?

Zdenko

st 2. 10. 2024 o 7:35 L ht <lhta...@gmail.com> napísal(a):

--
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 tesseract-oc...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/afe1dc25-cebc-43c9-94bc-dcabe6d43c52n%40googlegroups.com.

[L ht]

Yes, I did. 

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAJbzG8zfnB%3Dhk2nwBB-DN%3DHUpG5ibmrceHiV6-gVyntWr_GCEA%40mail.gmail.com.

[Zdenko Podobny]

Great!  So you know what preprocessing steps you need to take to get correct results.

>tesseract improved.png - --psm 7
SIGNIN

Zdenko

so 5. 10. 2024 o 20:32 L ht <lhta...@gmail.com> napísal(a):

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CANmU3o9dP29QKkRkXKMUO%2BXCpCpK-vW9SwpES%2BSwcLYkXEAvkQ%40mail.gmail.com.

[L ht]

Hi Zdenko, 

Thanks

I managed to get the OCR result by cropping out all the white edges from the "SIGN IN" image. Interestingly, I didn’t need to crop the white edges from the "SIGN IN WITH FACEBOOK" image to achieve a successful result. Although both images appear quite similar—white text on a dark background, with some white edges—the results differed between them. Could you please provide more details about how you handled this, so I can better manage similar cases in the future?  

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAJbzG8x18pUsTRt4S%3D7ADV7Nee%2BWvAYjMKFY4UC7Swj0XKzOSA%40mail.gmail.com.

[Zdenko Podobny]

Hi,

I guess the problem is:

• the word is short 
• surrounded by graphics elements (button). 

tesseract does a little image preprocessing (a user is responsible for preprocessing ;-) ) e.g. converting image to binary image(black&white) and it tries to ignore pictures in the provided image... 

Based on your images: you can remove a button with the following Python code:

import cv2

import numpy as np

img = cv2.imread('Images/notDetected.png', cv2.IMREAD_GRAYSCALE)
_, im_th = cv2.threshold(img, 200, 255, cv2.THRESH_BINARY_INV)
im_floodfill = im_th.copy()
h, w = im_th.shape[:2]
mask = np.zeros((h+2, w+2), np.uint8)
cv2.floodFill(im_floodfill, mask, (0, 0), (255, 255, 255))
cv2.imwrite('improved.png', im_floodfill)

Zdenko

po 7. 10. 2024 o 20:47 L ht <lhta...@gmail.com> napísal(a):

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CANmU3o9L2_WUSu6V53dSdpvqv8gkNXqhK0FYghXA_y1ajobBEw%40mail.gmail.com.