使用 openCV 从轮廓构建嵌套掩码

Building nested mask from contours with openCV

我想根据我绘制的轮廓构建一个嵌套蒙版(带孔的蒙版)。

输入轮廓图像附加到此消息 - 称为 contours.png -,这是我用来构建嵌套掩码的代码。

import cv2
import numpy as np
import matplotlib.pyplot as plt

test_im = cv2.imread("contours.png")
im_gray = cv2.cvtColor(test_im, cv2.COLOR_RGB2GRAY)

# find contours and hierarchy with OpenCV 
cnts, hierachy = cv2.findContours(im_gray, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
cnts = np.array(cnts)

mask = np.zeros_like(test_im)

# draw nested mask from contours using cv2.fillPoly
for i, cnt in enumerate(cnts):

    # look for external contours
    if hierachy[0][i][3] == -1:
        cnt = cnt.reshape((cnt.shape[0], 2))
        # fill the external contour entirely
        cv2.fillPoly(mask, [cnt], 255)

        # look for grandchild contours to fill them with zeros (and have a nested mask as output)
        child_ix = hierachy[0][i][2]
        same_level_ix = hierachy[0][child_ix][0]

        # for an akward reason, the extrenal contour has two child contours
        # (should get only one in my understanding)
        if same_level_ix == -1:
            grandchild_ix = hierachy[0][child_ix][2]
        else:
            child_ix = hierachy[0][child_ix][2]
            grandchild_ix = hierachy[0][same_level_ix][2]

        if grandchild_ix != -1:
            cnt = cnts[grandchild_ix]
            cnt = cnt.reshape((cnt.shape[0], 2))
            cv2.fillPoly(mask, [cnt], 0)
            same_level_ix = hierachy[0][grandchild_ix][0]

            while same_level_ix != -1:
                cnt = cnts[same_level_ix]
                cnt = cnt.reshape((cnt.shape[0], 2))
                cv2.fillPoly(mask, [cnt], 0)
                same_level_ix = hierachy[0][same_level_ix][0]

即使它适用于此示例,我的代码似乎也不是很可靠。另外,我发现外部轮廓有两个子轮廓,这对我来说很奇怪:在我的理解中应该只有一个。

你有更好的解决办法吗?

感谢您的帮助,祝您愉快!

contours.png

desired_output

根据hierarchy设置,可以做2个mask,然后相减得到结果。第一个掩码是通过填充最外层的轮廓来完成的,第二个掩码是通过填充最内层的轮廓来完成的:

这里是抽取轮廓和填充轮廓的必要设置(c++中的代码,但设置等同于python):

Mat img__1, img__2,img__ = imread("E:/img.jpg", 0);

threshold(img__, img__1, 0, 255, THRESH_BINARY);

vector<vector<Point>> contours;
vector< Vec4i > hierarchy;

findContours(img__1, contours, hierarchy, RETR_TREE, CHAIN_APPROX_NONE);

Mat tmp = Mat::zeros(img__1.size(), CV_8U);
Mat tmp2 = Mat::zeros(img__1.size(), CV_8U);

for (size_t i = 0; i < contours.size(); i++)
    if (hierarchy[i][3]<0)
        drawContours(tmp, contours, i, Scalar(255, 255, 255), -1); # first mask

for (size_t i = 0; i < contours.size(); i++)
    if (hierarchy[i][2]<0 && hierarchy[i][3]>-1)
        drawContours(tmp2, contours, i, Scalar(255, 255, 255), -1); # second mask

imshow("img", img__1);
imshow("first_mask", tmp);
imshow("second_mask", tmp2);

tmp = tmp - tmp2;  # subtracting the two masks to remove the holes

imshow("final image", tmp);
waitKey(0);