I'm deskewing an image containing a number using Projection profile based skew estimation algorithm and extracting it through OCR.

In order to calculate the correct skew angle, we compare the maximum difference between peaks and using this skew angle, thus rotate the image using the correct angle to correct the skew.

Each image (which has a single number) has prefixed background and foreground (number) color:

• If the first pixel of the background is black, the foreground is dark gray.
• If the first pixel of the background is white, the foreground is dark white.

Here are some sample images:

which all of them get successfully deskewed to these:

After it's been deskewed, I've tried with no luck to improve image quality to let OCR (PyTesseract) recognize the numbers.

import cv2
import numpy as np
import scipy.ndimage
import pytesseract
from PIL import Image, ImageEnhance, ImageFilter


from scipy.ndimage import interpolation as inter

def correct_skew(image, delta=6, limit=150):
    def determine_score(arr, angle):
        data = inter.rotate(arr, angle, reshape=False, order=0)
        histogram = np.sum(data, axis=1)
        score = np.sum((histogram[1:] - histogram[:-1]) ** 2)
        return histogram, score
 
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)                              
    blur = cv2.medianBlur(gray, 21)                                             
    thresh = cv2.threshold(blur, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]


    scores = []
    angles = np.arange(-limit, limit + delta, delta)
    for angle in angles:
        histogram, score = determine_score(thresh, angle)
        scores.append(score)

    best_angle = angles[scores.index(max(scores))] + 90

    (h, w) = image.shape[:2]
    center = (w // 2, h // 2)
    M = cv2.getRotationMatrix2D(center, best_angle, 1.0)
    rotated = cv2.warpAffine(image, M, (w, h), flags=cv2.INTER_CUBIC)


    ## Image processing to improve OCR accuracy
    """
    #rotated = cv2.medianBlur(rotated, 20)
    
    rotated = rotated.astype(np.float) / 255.

    # Calculate channel K:
    rotated = 1 - np.max(rotated, axis=2)

    # Convert back to uint 8:
    rotated = (255 * rotated).astype(np.uint8)

    binaryThresh = 190
    _, binaryImage = cv2.threshold(rotated, binaryThresh, 255, cv2.THRESH_BINARY)

    binaryThresh = 190
    _, binaryImage = cv2.threshold(rotated, binaryThresh, 255, cv2.THRESH_BINARY)

    # Use a little bit of morphology to clean the mask:
    # Set kernel (structuring element) size:
    kernelSize = 3
    # Set morph operation iterations:
    opIterations = 2
    # Get the structuring element:
    morphKernel = cv2.getStructuringElement(cv2.MORPH_RECT, (kernelSize, kernelSize))
    # Perform closing:
    rotated = cv2.morphologyEx(binaryImage, cv2.MORPH_CLOSE, morphKernel, None, None, opIterations, cv2.BORDER_REFLECT101)
    """

    return best_angle, rotated

if __name__ == '__main__':
    image = cv2.imread('number5.jpg')
    if image[0][0][0] > 128: image = cv2.bitwise_not(image)
    angle, rotated = correct_skew(image)
    print(angle)
    cv2.imshow('rotated', rotated)
    cv2.imwrite('rotated.png', rotated)
    pytesseract.pytesseract.tesseract_cmd = r'C:\\Program Files\\Tesseract-OCR\\tesseract.exe'

    text = pytesseract.image_to_string(rotated, config="-c tessedit_char_whitelist=0123456789")
    print("number:", text)
    cv2.waitKey()

This code (PyTesseract) recognizes the first and the second number, while not the others. Why?

I've already tried some tweaks concerning:

• Rescaling
• Binarisation
• Noise Removal
• Dilation / Erosion
• Rotation / Deskewing
• Borders
• Transparency / Alpha channel