deskewing.py

import numpy as np
import math
import cv2


def compute_skew(file_name):
    # load in grayscale:
    src = cv2.imread(file_name, 0)
    height, width = src.shape[0:2]

    # invert the colors of our image:
    cv2.bitwise_not(src, src)

    # Hough transform:
    minLineLength = width / 2.0
    maxLineGap = 20
    lines = cv2.HoughLinesP(src, 1, np.pi / 180, 100, minLineLength, maxLineGap)

    # calculate the angle between each line and the horizontal line:
    angle = 0.0
    nb_lines = len(lines)

    for line in lines:
        angle += math.atan2(line[0][3] * 1.0 - line[0][1] * 1.0, line[0][2] * 1.0 - line[0][0] * 1.0);

    angle /= nb_lines * 1.0

    return angle * 180.0 / np.pi


def deskew(file_name, angle):
    # load in grayscale:
    img = cv2.imread(file_name, 0)

    # invert the colors of our image:
    cv2.bitwise_not(img, img)

    # compute the minimum bounding box:
    non_zero_pixels = cv2.findNonZero(img)
    center, wh, theta = cv2.minAreaRect(non_zero_pixels)

    root_mat = cv2.getRotationMatrix2D(center, angle, 1)
    rows, cols = img.shape
    rotated = cv2.warpAffine(img, root_mat, (cols, rows), flags=cv2.INTER_CUBIC)

    # Border removing:
    sizex = np.int0(wh[0])
    sizey = np.int0(wh[1])
    print(theta)
    if theta > -45:
        temp = sizex
        sizex = sizey
        sizey = temp
    return cv2.getRectSubPix(rotated, (sizey, sizex), center)


file_path = 'D:\\September\\3 PAN Card OCR - WIP\\pranjal1.png'
angel = compute_skew(file_path)
dst = deskew(file_path, angel)
cv2.imshow("Result", dst)
cv2.waitKey(0)