How to read all the text from the image?

Question

I am using pytesseract to read text from an image..

I am facing the following challenges:

• It is not extracting all the text.
• All the text read is outputting u'\x0c'
• It is not forming proper text

image_to_data was able to convert a few but most of it was not right. I am attaching the code below

fn = 'testimgnum.png'
image = cv2.imread(fn)

gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
gray = cv2.GaussianBlur(gray, (5, 5), 0)    
gray = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]

'''
gray = cv2.Canny(gray, 100, 200)
gray = cv2.dilate(gray, None, iterations=1)
gray = cv2.erode(gray, None, iterations=1)
'''

cv2.imwrite('nt1.png', gray)

cnts = cv2.findContours(gray.copy(), cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE)
cnts = imutils.grab_contours(cnts)

(cnts, _) = contours.sort_contours(cnts)
orig = gray.copy()
arr_values = []
ctr = 0
#filter the boxes
custom_config = r'--oem 3 --psm 10000'
k = 0
for c in cnts:
    k=k+1
    #print("c:",c)
    try:
        text_val=''
        box = cv2.minAreaRect(c)
        box = cv2.cv.BoxPoints(box) if imutils.is_cv2() else cv2.boxPoints(box)
        box = np.array(box, dtype="int")
        tl, tr, br, bl = box
        #print(box)
        width = abs(br[0]-tl[0])
        height = abs(tl[1]-br[1])
        x1= min(tl[0],br[0])
        x2= max(tl[0],br[0])
        y1=max(tl[1],br[1])
        y2 = min(tl[1],br[1])
        print(x1,x2,y1,y2, height, width)    
        crop_img = gray[y2-6:y1+6, x1-4:x2+4]
        cv2.imwrite('{0}.png'.format(k), crop_img)

        cv2.imshow("Image", crop_img)
        cv2.waitKey(0)
        #crop_img1 = cv2.resize(crop_img, None, fx=6, fy=6, interpolation=cv2.INTER_CUBIC)
        text_val = pytesseract.image_to_string(crop_img, lang='eng',config='--oem 3 -c tessedit_char_whitelist=0123456789')#, config=custom_config,
        print("text_val:", text_val)
        arr_values.append([x1,x2,y1,y2,width,height,text_val])

    except Exception as e:
        print("e:", e)
        ctr+=1
        pass

I am using tesseract 4 and python 2.7. I had a working model for a similar problem. I was using an older version of tesseract.

Answer 1

That's my approach:

1. Mask white-ish parts in image to find the numbers, and dilate to merge single digits.
2. Find contours, and get bounding rectangles.
3. Use the saturation channel from the HSV converted ROI for the actual OCR, also using the -psm 6 option.

That'd be the full code:

import cv2
import pytesseract

# Read image
img = cv2.imread('KbZMp.png')

# Mask white-ish parts
mask = (img > 248).all(axis=2).astype(img.dtype) * 255
mask = cv2.dilate(mask, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (11, 11)))

# Find contours w.r.t. the OpenCV version; retrieve bounding rectangles
cnts = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
cnts = cnts[0] if len(cnts) == 2 else cnts[1]
rects = [cv2.boundingRect(cnt) for cnt in cnts]

# Just for visualization purposes
out = img.copy()

# Iterate bounding rectangles, and OCR
for x, y, w, h in rects:

    # Use saturation channel of ROI for actual OCR
    sat = cv2.cvtColor(img[y:y+h, x:x+w], cv2.COLOR_BGR2HSV)[..., 1]

    # Pytesseract with -psm 6
    text = pytesseract.image_to_string(sat, config='--psm 6')\
        .replace('\n', '').replace('\f', '')

    # Just for visualization purposes
    out = cv2.rectangle(out, (x, y), (x+w, y+h), (0, 255, 0), 2)
    out = cv2.putText(out, text, (x, y-5),
                      cv2.FONT_HERSHEY_COMPLEX_SMALL, 1, (0, 0, 255))

# Just for visualization purposes
cv2.imshow('out', out)
cv2.waitKey(0)

And, that's a visualization of the results:

Caveat: I use a special version of Tesseract from the Mannheim University Library.

----------------------------------------
System information
----------------------------------------
Platform:      Windows-10-10.0.19041-SP0
Python:        3.9.1
PyCharm:       2021.1.1
OpenCV:        4.5.2
pytesseract:   5.0.0-alpha.20201127
----------------------------------------