具有细化和命中或未命中实现的骨架化从未停止

Question

我正在尝试使用https://homepages.inf.ed.ac.uk/rbf/HIPR2/thin.htm 中所述的细化/命中或未命中方法来实现骨架化算法。

我想提取以下图像的骨架，我认为这是一个很好的测试图像：

https://github.com/TiagoB936/Image-Processing-and-Computer-Vision/blob/master/Skeletonization/x.png

问题是我的代码无法停止运行，我没有发现问题。

我的代码执行以下步骤：

• 读取图像并将其二值化

• 创建 2 个主内核：

  [[0,0,0]

  [-1,1,-1]

  [1,1,1]]

  和

  [[-1,0,0]

  [1,1,0]

  [-1,1,1]]

• 再创建6个内核，每个内核旋转90度主内核

• 应用命中或未命中创建 8 个输出图像
• 将这些图像相加并减去输入图像的结果生成输出图像
• 验证输出图像是否等于输入图像
• 输入图像变成输出图像
• 在输入和输出图像之间没有差异时重复

这是一些测试代码：

import cv2 as cv
import numpy as np

input_image = cv.imread('x.png',0)
_,input_image = cv.threshold(input_image,60,255,cv.THRESH_BINARY)

kernel1 = np.array((
        [0, 0, 0],
        [-1, 1, -1],
        [1, 1, 1]), dtype="int")
kernel2 = np.array((
        [-1, 0, 0],
        [1, 1, 0],
        [-1, 1, -1]), dtype="int")
kernel3 = np.rot90(kernel1)
kernel4 = np.rot90(np.rot90(kernel1))
kernel5 = np.rot90(np.rot90(np.rot90(kernel1)))
kernel6 = np.rot90(kernel2)
kernel7 = np.rot90(np.rot90(kernel2))
kernel8 = np.rot90(np.rot90(np.rot90(kernel2)))


output_image_1 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel1)
output_image_2 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel2)
output_image_3 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel3)
output_image_4 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel4)
output_image_5 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel5)
output_image_6 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel6)
output_image_7 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel7)
output_image_8 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel8)
hit_miss = output_image_1 + output_image_2 + output_image_3 + output_image_4 + output_image_5 + output_image_6 + output_image_7 + output_image_8
output_image = input_image - hit_miss

while not np.array_equal(output_image, input_image):
    input_image = output_image
    output_image_1 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel1)
    output_image_2 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel2)
    output_image_3 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel3)
    output_image_4 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel4)
    output_image_5 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel5)
    output_image_6 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel6)
    output_image_7 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel7)
    output_image_8 = cv.morphologyEx(input_image, cv.MORPH_HITMISS, kernel8)
    hit_miss = output_image_1 + output_image_2 + output_image_3 + output_image_4 + output_image_5 + output_image_6 + output_image_7 + output_image_8
    output_image = input_image - hit_miss

    cv.imshow("hit_miss", hit_miss)
    cv.imshow("out", output_image)  
    cv.waitKey(0)
    cv.destroyAllWindows()

我希望从我的输入中得到一个薄的 X 图像，但代码无法停止运行。我错过了什么吗？

Answer 1

您想在应用所有内核之后使用逻辑或，而不是仅仅添加它们。

import cv2 as cv
import numpy as np

img = cv.imread('x.png',0)
_,img = cv.threshold(img,60,255,cv.THRESH_BINARY)

cv.imshow("input", img)  
cv.waitKey(0)
cv.destroyAllWindows()

kernel1 = np.array((
        [[-1, -1, -1],
        [0, 1, 0],
        [1, 1, 1]]))
kernel2 = np.array((
        [[0, -1, -1],
        [1, 1, -1],
        [0, 1, 0]]))

kernel3 = np.rot90(kernel1)
kernel4 = np.rot90(kernel2)
kernel5 = np.rot90(np.rot90(kernel1))
kernel6 = np.rot90(np.rot90(kernel2))
kernel7 = np.rot90(np.rot90(np.rot90(kernel1)))
kernel8 = np.rot90(np.rot90(np.rot90(kernel2)))
skel = np.zeros(img.shape,np.uint8)
mask = np.zeros(img.shape,np.uint8)
kernels = [kernel1, kernel2, kernel3, kernel4, kernel5, kernel6, kernel7, kernel8]

while (1):
    mask = np.zeros(img.shape,np.uint8)

    for kernel in kernels:
        out_image = cv.morphologyEx(img, cv.MORPH_HITMISS, kernel)
        mask = cv.bitwise_or(out_image, mask)

    img = img - mask

    if (np.array_equal(img, skel)):
        break

    skel = img

cv.imshow("out", img)  
cv.waitKey(0)
cv.destroyAllWindows()