为什么 pytesseract 无法识别此图像

Question

I am testing the pytesseract OCR on this image

but the result is always 30770.0 but I would have wanted this number: 997,70 FYI: this image has already been transformed:

img = img.convert('L')  # greyscale
img = img.resize((img.size[0] * 3, img.size[1] * 3), 1)
img = ImageEnhance.Contrast(img).enhance(5.0)
img = ImageOps.equalize(img)
img = ImageOps.invert(img)

Below the code:

pytesseract.pytesseract.tesseract_cmd = 'C:\\Program Files\\Tesseract-OCR\\tesseract.exe'
img2 = "full_snap__1609584655.png"
numStr2 = pytesseract.image_to_string(img2, lang='eng',config='--psm 10 --oem 1 digits -c tessedit_char_whitelist=0123456789')
print('997,70 :',float(numStr2))

I have already tried to adjust the --psm parameter of the pytesseract function image_to_string but it doesn't work.

Thank you for your help

Answer 1

My solution to the problem is morphological-transformation.

If you apply erosion

the thickness or size of the foreground object decreases or simply white region decreases in the image. It is useful for removing small white noises (as we have seen in colorspace chapter), detach two connected objects etc.

erd = cv2.erode(gry, None, iterations=1)

Result:

Now, if you read it:

print(pytesseract.image_to_string(erd))

Result:

997 70€

Code:

---

import cv2
import pytesseract

img = cv2.imread("R83OY.png")
h, w, c = img.shape
gry = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
erd = cv2.erode(gry, None, iterations=1)
print(pytesseract.image_to_string(erd))

Possible Question: Why do you set kernel to None?

If you initialize a kernel (ie (5, 5) ) and apply it to the image, result will be:

As you can see, applying kernel is not improving the result.