Extract text from image using tessaract and opencv

Question

screenshot.png:

modified_image.png:

I am trying to extract text from an image but seems however I do it tessaract gives me some random values even though I think I have processed the image to a very good format. I am only after the white text and want to disregard the red text.

import cv2 as cv
import pytesseract
from PIL import Image

image = cv.imread("screenshot.png", cv.IMREAD_GRAYSCALE)

ret, modified_image = cv.threshold(image, 120, 255, cv.THRESH_BINARY_INV + cv.THRESH_OTSU)

modified_image= cv.resize(modified_image, None, fx=2, fy=2, interpolation=cv.INTER_CUBIC)

#cv.imshow("image", image)
#cv.imshow("modified_image", modified_image)
cv.imwrite("modified_image.png", modified_image)

pytesseract.pytesseract.tesseract_cmd = r'C:\Tesseract-OCR\tesseract.exe'

text = pytesseract.image_to_string(Image.open('modified_image.png'), config="--psm 6 --oem 3", lang="eng")

print(f'Text: {text}')

This will incorrectly print "CWS-1Y" instead of "CW9-1Y". From what I have understood the font in use is Shentox but seems like quite the task to train tessaract on it from what I could find online

@user24714692 All images are like that, the format of letters/digits can be different though (they come from a game) — Andreas Ellsen
– Andreas Ellsen, Commented May 14, 2024 at 18:13
Send me your ISK and I will double it! Jokes aside, have you tried not resizing the image, or using different scale factors, or different interpolation, or higher threshold? — GSazheniuk
– GSazheniuk, Commented May 14, 2024 at 18:54
I'm out of idea for using Tesseract. Can't do better than this. Instead, you can clean, segment your images, and write grid algorithms to find these letters (character by char) with a much higher accuracy. (Fails to recognize between S and 9 which have similar patterns). — Aicody
– Aicody, Commented May 14, 2024 at 19:05
@GSazheniuk haha :) Yeah been trying different things for hours — Andreas Ellsen
– Andreas Ellsen, Commented May 14, 2024 at 19:34

divine architect · Accepted Answer · 2024-05-14 23:45:53Z

2

Tesseract, while being "good" is not the best. For critical applications I tend to use EasyOCR so maybe give that a shot.

It's free and opensource as well.

Here's an example:

import easyocr
reader = easyocr.Reader(['en']) # this needs to run only once to load the model into memory



result = reader.readtext('chinese.jpg')
print(result)

edited May 14, 2024 at 23:45

answered May 14, 2024 at 22:14

divine architect

687 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Andreas Ellsen Over a year ago

I have tried it also with similar results so not optimal either

user898678 · Accepted Answer · 2024-05-16 12:17:32Z

0

Read the docs: https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md
Adjust image according docs:
- Resize image to meet optimal letter size
- Invert image (dark letters on bright background
- (optional) convert to grayscale (or binarize)

tesseract modified_image_based_on_docs.png - --psm 6
CW9-1Y

answered May 16, 2024 at 12:17

user898678

3,3832 gold badges22 silver badges20 bronze badges

Collectives™ on Stack Overflow

Extract text from image using tessaract and opencv

2 Answers 2

1 Comment

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related