opencv 如何校正扫描图像中支票/文本的图像歪斜？

sdnqo3pr 于 2023-01-09 发布在其他

关注(0)|答案(1)|浏览(187)

这是我需要校正

的图像我不想在这里运行对象检测，我尝试了以下脚本，但不起作用，下一步该怎么办？

import math
from typing import Tuple, Union

import cv2
import numpy as np

from deskew import determine_skew

def rotate(
        image: np.ndarray, angle: float, background: Union[int, Tuple[int, int, int]]
) -> np.ndarray:
    old_width, old_height = image.shape[:2]
    angle_radian = math.radians(angle)
    width = abs(np.sin(angle_radian) * old_height) + abs(np.cos(angle_radian) * old_width)
    height = abs(np.sin(angle_radian) * old_width) + abs(np.cos(angle_radian) * old_height)

    image_center = tuple(np.array(image.shape[1::-1]) / 2)
    rot_mat = cv2.getRotationMatrix2D(image_center, angle, 1.0)
    rot_mat[1, 2] += (width - old_width) / 2
    rot_mat[0, 2] += (height - old_height) / 2
    return cv2.warpAffine(image, rot_mat, (int(round(height)), int(round(width))), borderValue=background)

image = cv2.imread('/content/img.jpg')
grayscale = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
angle = determine_skew(grayscale)
rotated = rotate(image

, angle, (0, 0, 1))
cv2_imshow(rotated)

还有这个简单的版本

import numpy as np
from skimage import io
from skimage.color import rgb2gray
from skimage.transform import rotate

from deskew import determine_skew

image = io.imread('/content/img.jpg')
grayscale = rgb2gray(image)
angle = determine_skew(grayscale)
rotated = rotate(image, angle, resize=True) * 255
# io.imsave('output.png', rotated.astype(np.uint8))
cv2_imshow(rotated.astype(np.uint8))

opencv

来源：https://stackoverflow.com/questions/75018001/how-do-i-deskew-the-image-of-a-cheque-text-in-a-scanned-image

1条答案

按热度按时间

bvjxkvbb1#

这是我在我的谷歌colab上运行的脚本，用来消除它的歪斜

import cv2
import numpy as np
from google.colab.patches import cv2_imshow
from skimage import io

def deskew(im, max_skew=10):
    height, width , _= im.shape

    # Create a grayscale image and denoise it
    im_gs = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
    im_gs = cv2.fastNlMeansDenoising(im_gs, h=3)

    # Create an inverted B&W copy using Otsu (automatic) thresholding
    im_bw = cv2.threshold(im_gs, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1]

    # Detect lines in this image. Parameters here mostly arrived at by trial and error.
    lines = cv2.HoughLinesP(
        im_bw, 1, np.pi / 180, 200, minLineLength=width / 12, maxLineGap=width / 150
    )

    # Collect the angles of these lines (in radians)
    angles = []
    for line in lines:
        x1, y1, x2, y2 = line[0]
        angles.append(np.arctan2(y2 - y1, x2 - x1))

    # If the majority of our lines are vertical, this is probably a landscape image
    landscape = np.sum([abs(angle) > np.pi / 4 for angle in angles]) > len(angles) / 2

    # Filter the angles to remove outliers based on max_skew
    if landscape:
        angles = [
            angle
            for angle in angles
            if np.deg2rad(90 - max_skew) < abs(angle) < np.deg2rad(90 + max_skew)
        ]
    else:
        angles = [angle for angle in angles if abs(angle) < np.deg2rad(max_skew)]

    if len(angles) < 5:
        # Insufficient data to deskew
        return im

    # Average the angles to a degree offset
    angle_deg = np.rad2deg(np.median(angles))

    # If this is landscape image, rotate the entire canvas appropriately
    if landscape:
        if angle_deg < 0:
            im = cv2.rotate(im, cv2.ROTATE_90_CLOCKWISE)
            angle_deg += 90
        elif angle_deg > 0:
            im = cv2.rotate(im, cv2.ROTATE_90_COUNTERCLOCKWISE)
            angle_deg -= 90

    # Rotate the image by the residual offset
    M = cv2.getRotationMatrix2D((width / 2, height / 2), angle_deg, 1)
    im = cv2.warpAffine(im, M, (width, height), borderMode=cv2.BORDER_REPLICATE)
    return im 



pt = "/content/126016027_AO_BIZ-0000320520-Process_IP_Cheque.jpg"
image = io.imread(pt)
img = deskew(image) 
cv2_imshow(img)

赞(0）回复(0）举报 2023-01-09

我来回答

opencv 如何校正扫描图像中支票/文本的图像歪斜？

1条答案

相关问题

热门标签

最新问答