Pytesseract only gets a part of a text from the image

Question

I am trying to use the pytesseract to detect the text from this image:

Specifically, I care more about detecting Commercial break in progress. I used the following code to achieve this:

from PIL import Image
import pytesseract
print(pytesseract.image_to_string(Image.open('/home/me/Desktop/dataset/my-dataset/Apex-Legends/loustreams_001.jpg')))

However, it returns:

nercial break in progress

I know I shouldn't expect SoTA result from one line of code but the text is very visible. How can I improve this?

Answer 1

You can use image_to_data to get the "Commercial break in progress" string.

1. Convert image to rgb

2. Get output as a dictionary

3. If detected text pass the threshold confidence level, display.

Code:

import cv2
from pytesseract import *

img = cv2.imread("6cNav.jpg")
rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
res = image_to_data(rgb, output_type=Output.DICT)

out_txt = ""

for i in range(0, len(res["text"])):
    x = res["left"][i]
    y = res["top"][i]
    w = res["width"][i]
    h = res["height"][i]

    txt = res["text"][i]
    cnf = int(res["conf"][i])

    if cnf > 95:
        text = "".join(txt).strip()
        cv2.rectangle(img,
                      (x, y),
                      (x + w, y + h),
                      (0, 0, 255), 2)
        cv2.putText(img,
                    text,
                    (x, y - 10),
                    cv2.FONT_HERSHEY_SIMPLEX,
                    1.2, (0, 255, 255), 3)
        out_txt += " " + text

print(out_txt)
cv2.imshow("img", img)
cv2.waitKey(0)

Output:

Commercial break in progress

Please note that my pytesseract version is 4.1.1

Updated Code

---

For the all the images above, you can apply adaptive-threshold

• (1st and the 3rd image is also similar to the above)

the result will be:

output 1:
Commercial loreak in progress

output 2:
Commercial break in progress

output 3:
Commercial break in progress

Code:

import cv2
import pytesseract

img1 = cv2.imread("6cNav.jpg")
img2 = cv2.imread("IKHKa.jpg")
img3 = cv2.imread("VTJse.jpg")

img_lst = [img1, img2, img3]

for i, img in enumerate(img_lst):
    gry = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    thr = cv2.adaptiveThreshold(gry, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
                                cv2.THRESH_BINARY_INV, 11, 2)

    res = pytesseract.image_to_string(thr, config="--psm 6")
    print("output {}:".format(i+1))
    print(res)