Hello,
I managed to implement a dynamic parsing to get rid of OSD issues i had.
However i'm blocking on recognizing single uppercase letter, i tried many
different configurations for preprocessing but i can't get to find the
right one, even with PSM set to 10, i don't really know what i could try.
Any help is appreciated.
Here is code snippet for testing with pictures attached :
import cv2
import os
import pytesseract
import numpy as np
pytesseract.pytesseract.tesseract_cmd = r'C:\Program Files\Tesseract-OCR\t
esseract.exe'
for pic in ["e.png","d-.png","d.png"]:
img=cv2.imread(pic)
#Preprocessing
img = cv2.resize(img, (70, 90), interpolation=cv2.INTER_NEAREST)
norm_img = np.zeros((img.shape[0], img.shape[1]))
img = cv2.normalize(img, norm_img, 0, 255, cv2.NORM_MINMAX)
img = cv2.fastNlMeansDenoisingColored(img, None, 10, 10, 7, 15)
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
img = cv2.bitwise_not(img)
img = cv2.threshold(img,127,255,cv2.THRESH_BINARY) [1]
cv2.imwrite("processed-"+pic, img)
# Tesseract OCR
text = pytesseract.image_to_string(img, lang='eng', config='-c
tessedit_char_whitelist=\\ ABCDEF+- tessedit_char_blacklist=\\=!,*%^$°:.
--psm 10 -oem 3')
print(str(text).replace("\n", " "))
Le mercredi 7 février 2024 à 06:39:37 UTC+1, dev 313153 a écrit :
> Hello,
> I am very new to tesseract, as well as in image processing in general.
> I have screenshots from which i want to extract text for further
> processing, i played around with tesseract after checking the Improve
> Quality URL and was able to extract what i need (most of the time).
> For example, in attached screenshots, i want to extract names of the stats
> and the following letter together, but it doesn't always work.
> Sometime the letter isn't extracted, and sometime it is, but the OSD
> consider it belongs on an other level or row and it's output ahead or
> before the stats names when i use image_to_string.
> I also tried to play with oem and psm settings, without much improvements.
>
> I attached some example of image_to_string outputs for different pictures
> as well as images and the python code i'm using as testing bench.
>
> I am getting a bit desesperate, so i consider the following approaches :
> - training my own dataset for this need, having sufficient data shouldn't
> be an issue over time but i have zero experience on this kind of thing.
> - looking for the stats names coordinates, and then cropping the picture
> around it to make sure tesseract focusses on it and extract it properly
> (sounds like a chore code wise, but doable i think).
>
> Let me know what you think about it or if you have a improvements to
> suggest.
> Best Regards,
>
>
--
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 on the web visit
https://groups.google.com/d/msgid/tesseract-ocr/cd13256e-46f1-405a-842b-e2d781d22e4en%40googlegroups.com.
