需要帮助找到 Pytesseract 的正确配置才能识别此文本

Question

So I'm doing this script where I need to be able to recognize text from this type of pictures [1][2], namely the text "Curse of Binding" and "Looting I". I'm new to Pytesseract and C2V, so I really don't know how I should edit the pictures. The following code doesn't get me the result I want.

import pytesseract
import cv2

pytesseract.pytesseract.tesseract_cmd = "C:\\Users\\guilh\\AppData\\Local\\Programs\\Tesseract-OCR\\tesseract.exe"

image = cv2.imread('trades/1.png')

text = pytesseract.image_to_string(image, lang='mc')
print(result)

Thanks in advance!

[1] - https://i.stack.imgur.com/0fNhM.png

[2] - https://i.stack.imgur.com/J71v6.png

Edit: the lang "mc" is a custom font for the one in the pictures, from the game Minecraft.

Answer 1

This may be a far fetched question, but did you download the mc trained data from this link ?

If so, this training data has problems with certain characters and only really works well for numbers. Another important thing to do is try to cut out the background around the text.

I'm doing a similar project here with a few differences. (using tesserocr as its faster for video/large amounts of images) (reading the f3 debug menu which guarantees white text)

If you have a look at process_image, it takes the image, cuts out all non-gray pixels, then applies cv2.threshold(im_arr,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)

I'm also using a crop to the first column that actually has text after applying all effects to minimize large amounts of white space at the start.

Instead of checking for gray pixels, you could try using

# Check out hsv masking/filtering in opencv documentation
image = cv2.inRange((h_min,s_min,v_min), (h_max, s_max, v_max), image)
ret3, image = cv2.threshold(image,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)

def process_image(im, crop_to_activity=False):
    """
    Converts the image to a numpy array, then applies preprocessing.
    """
    im_arr = np.array(im)
        
    height, width, depth = im_arr.shape

    for i in range(height):
        for j in range(width):
            r, g, b = im_arr[i][j]

            r = (r + 150) / 2
            g = (g + 150) / 2
            b = (b + 150) / 2

            mean = (r + g + b) / 3
            diffr = abs(mean - r)
            diffg = abs(mean - g)
            diffb = abs(mean - b)

            maxdev = 2

            if (diffr + diffg + diffb) > maxdev:
                im_arr[i][j][0] = 0
                im_arr[i][j][1] = 0
                im_arr[i][j][2] = 0
            


    im_arr = cv2.cvtColor(im_arr, cv2.COLOR_BGR2GRAY)

        #cap_arr = cv2.threshold(cap_arr,127,255,cv2.THRESH_BINARY)
    
    # Otsu's thresholding after Gaussian filtering
    #blur = cv2.GaussianBlur(cap_arr,(3,3),0)
    ret3, im_arr = cv2.threshold(im_arr,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)


    if crop_to_activity:
        last_column = -1
        for j in range(width):
            for i in range(height):
                v = im_arr[i][j]

                if v != 0:
                    last_column = j
                    break
            if last_column != -1:
                break
        
        last_column = max(0, last_column-3)
        im_arr = im_arr[:, last_column:]

    return im_arr

Answer 2

I would like to suggest inRange Thresholding

If you apply:

1st Image	2nd Image

If you set the page segmentation result to 6:

1st Image	2nd Image
Enchanted Book	Enchanted Book
Curse of Binding

In the 2nd image the looting-I is missing so we need to set different values:

2nd Image	Result
	dese eee _, Looting I

The global thresholding won't work on those images.

Adaptive thresholding is not more successful than inRange thresholding:

1st Image	2nd Image

As a result, you can find the ideal result by changing the values.

InRange threshold code:

import cv2
import pytesseract
from numpy import array

image_list = ["0fNhM.png", "J71v6.png"]

for image_name in image_list:
    bgr_image = cv2.imread(image_name)
    hsv_image = cv2.cv2.cvtColor(bgr_image, cv2.COLOR_BGR2HSV)
    mask = cv2.inRange(hsv_image, array([0, 158, 233]), array([46, 255, 255]))
    kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 3))
    dilate = cv2.dilate(mask, kernel, iterations=1)
    thresh = 255 - cv2.bitwise_and(dilate, mask)
    txt = pytesseract.image_to_string(thresh, config='--psm 6')
    print(txt)