[FIXED] Unable to get all numbers pytesseract

Issue

Hello i am new to python and i am trying to get numbers from an image using pytesseract and cv2, whenever i run my code its always missing a number or works well, for instance, it will miss the number 8 in the below image, also please note the numbers in the input could be random or empty as its a date input but a full date length i expect is 8, any help on getting accuracy would be really helpful.

15082022 there are 8 digits.

Thanks.

Code

img1 = cv2.imread("image.tiff")
gry1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
(h, w) = gry1.shape[:2]
gry1 = cv2.resize(gry1, (w*3, h*3))

thr1 = cv2.threshold(gry1, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
thr1 = cv2.GaussianBlur(thr1, (3, 3), 0)
txt1 = pytesseract.image_to_string(
    thr1, config=f"--oem 3 --psm 11 outputbase digits tessedit_char_whitelist=0123456789")

print("".join(t for t in txt1 if t.isalnum()))

cv2.imshow("thr1", thr1)
cv2.waitKey(0)

Output

1502022

Expected Output

15082022

Image example The image could contain text and might not, this depends on how the screenshot system did its work.

Solution

After playing around with tesseract and OpenCV I found several ways of how I could improve accuracy. However without proper benchmarking with a large dataset, I am not sure any answer would be helpful.

My approach was to:

• scale the image higher times instead of 3
• use cv2.erode() with a suitable kernel size.
• use --oem 3
• use --psm 7 ( Other options would also work, mode 7 tells tesseract to treat the image as a single text line)

I've also tried with several kernel sizes, and several scaling, and several iteration for the erode process, and there are many combinations that gives the wanted results.

If you find a way to remove the borders of the boxes the task would be easier

Answered By - Ouss