如何通过在框中使用计数来识别所有数字

Question

I'm learning openCV to recognize all infomation from transcript image. At No. column (leftmost), I use countours to get (x, y, x_max, y_max) and crop every box under gray image but can't recognize all digits in a few of boxes. I recognize only one of digit of this number. Here is my code:

import cv2

img = cv2.imread("original_img.jpg",0)
blur = cv2.GaussianBlur(img,(5,5),5)
thresh = cv2.adaptiveThreshold(blur,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY_INV,11,2)

horizal = thresh
vertical = thresh

scale_height = 30 
scale_long = 20

long = int(img.shape[1]/scale_long)
height = int(img.shape[0]/scale_height)

horizalStructure = cv2.getStructuringElement(cv2.MORPH_RECT, (long, 1))
horizal = cv2.erode(horizal, horizalStructure, (-1, -1))
horizal = cv2.dilate(horizal, horizalStructure, (-1, -1))

verticalStructure = cv2.getStructuringElement(cv2.MORPH_RECT, (1, height))
vertical = cv2.erode(vertical, verticalStructure, (-1, -1))
vertical = cv2.dilate(vertical, verticalStructure, (-1, -1))

mask = vertical + horizal
contours, hierarchy = cv2.findContours(mask,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE)

max = -1
for cnt in contours:
    x, y, w, h = cv2.boundingRect(cnt)
    if cv2.contourArea(cnt) > max:
        x_max, y_max, w_max, h_max = x, y, w, h
        max = cv2.contourArea(cnt)

table = img[y_max:y_max+h_max, x_max:x_max+w_max]
cropped_thresh_img = []
cropped_origin_img = []
countours_img = []

NUM_ROWS = 33
START_ROW = 1
for i in range(START_ROW, NUM_ROWS):
    thresh1 = thresh[y_max + round(i*h_max/NUM_ROWS):y_max + round((i+1)*h_max/NUM_ROWS), x_max + round(0*w_max/16):x_max +round(1*w_max/16)]
    contours_thresh1, hierarchy_thresh1 = cv2.findContours(thresh1, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    origin1 = img[y_max + round(i*h_max/NUM_ROWS):y_max + round((i+1)*h_max/NUM_ROWS), x_max + round(0*w_max/16):x_max +round(1*w_max/16)]

    cropped_thresh_img.append(thresh1)
    cropped_origin_img.append(origin1)
    countours_img.append(contours_thresh1)
cv2.imshow("thresh", cropped_thresh_img[20])
result = []
for i, countour_img in enumerate(countours_img):
    for cnt in countour_img:
        if cv2.contourArea(cnt) > 20:
             x,y,w,h = cv2.boundingRect(cnt)
             if x > cropped_origin_img[i].shape[1]*0.1 and x < cropped_origin_img[i].shape[1]*0.9:
                answer = cropped_origin_img[i][y:y+h, x:x+w]
                answer = cv2.threshold(answer, 80, 255, cv2.THRESH_BINARY_INV)[1]
                break
    result.append(answer)
cv2.imshow("No.21", result[20])

Image after contours:

Original image:

Answer 1

Why don't you take the part of the image you want to get data?

import cv2
import pytesseract

img = cv2.imread("gKGc9.png")
img = img[120:620, 25:45]  # (632, 445, 3)

Result

Part of the result:

You have already applied thresholding and find the contours

import cv2
import pytesseract

img = cv2.imread("gKGc9.png")
img = img[120:620, 25:45]  # (632, 445, 3)
cv2.imwrite("/Users/ahx/Desktop/res.png", img)
gry = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
thr = cv2.threshold(gry, 127, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
cnt = cv2.findContours(thr, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[0]

Now for each detected contour, you need to increase the scale. (ie 1.5 pixel)

for c in cnt:
    x, y, w, h = cv2.boundingRect(c)
    r = cv2.rectangle(img, (x, y), (x + w, y + h), (0, 255, 0), 2)
    crp = img[y:y + h, x:x + w]
    wth = int(crp.shape[1] * 150 / 100)
    hgt = int(crp.shape[0] * 150 / 100)
    crp = cv2.resize(crp, (wth, hgt), interpolation=cv2.INTER_AREA)

Now, you can use pytesseract to read the data

txt = pytesseract.image_to_string(crp, config="--psm 6 digits")

If text data is not empty, then display

    if txt != '\f':
        print(txt)
        cv2.imshow("crp", crp)
        cv2.waitKey(0)

Part of the results:

And in the console you should see:

31

26

25.

24

23

Code:

import cv2
import pytesseract

img = cv2.imread("gKGc9.png")
img = img[120:620, 25:45]  # (632, 445, 3)
cv2.imwrite("/Users/ahx/Desktop/res.png", img)
gry = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
thr = cv2.threshold(gry, 127, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
cnt = cv2.findContours(thr, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[0]
for c in cnt:
    x, y, w, h = cv2.boundingRect(c)
    r = cv2.rectangle(img, (x, y), (x + w, y + h), (0, 255, 0), 2)
    crp = img[y:y + h, x:x + w]
    wth = int(crp.shape[1] * 150 / 100)
    hgt = int(crp.shape[0] * 150 / 100)
    crp = cv2.resize(crp, (wth, hgt), interpolation=cv2.INTER_AREA)
    txt = pytesseract.image_to_string(crp, config="--psm 6 digits")
    if txt != '\f':
        print(txt)
        cv2.imshow("crp", crp)
        cv2.waitKey(0)

I installed the pytesseract via pip, and my pytesseract version is 4.1