二值分析 | OpenCV + skimage如何提取中心线

2020-10-27 10:56:02 浏览数 (1)

问题

前几天有个人问了我一个问题,问题是这样的,他有如下的一张二值图像:

怎么得到白色Blob中心线,他希望的效果如下:

显然OpenCV中常见的轮廓分析无法获得上面的中心红色线段,本质上这个问题是如何提取二值对象的骨架,提取骨架的方法在OpenCV的扩展模块中,另外skimage包也支持图像的骨架提取。这里就分别基于OpenCV扩展模块与skimage包来完成骨架提取,得到上述图示的中心线。

01

安装skimage与opencv扩展包

Python环境下安装skimage图像处理包与opencv计算机视觉包,只需要分别执行下面两行命令:

代码语言:javascript复制
pip install opencv-contrib-python
pip install skimage

导入使用

代码语言:javascript复制
from skimage import morphology
 import cv2 as cv

02

使用skimage实现骨架提取

有两个相关的函数实现二值图像的骨架提取,一个是基于距离变换实现的medial_axis方法;另外一个是基于thin的skeletonize骨架提取方法。两个方法的代码实现分别如下:

代码语言:javascript复制
def skeleton_demo(image):
    gray = cv.cvtColor(image, cv.COLOR_BGR2GRAY)
    ret, binary = cv.threshold(gray, 0, 255, cv.THRESH_BINARY | cv.THRESH_OTSU)
    binary[binary == 255] = 1
    skeleton0 = morphology.skeletonize(binary)
    skeleton = skeleton0.astype(np.uint8) * 255
    cv.imshow("skeleton", skeleton)
    cv.waitKey(0)
    cv.destroyAllWindows()


def medial_axis_demo(image):
    gray = cv.cvtColor(image, cv.COLOR_BGR2GRAY)
    ret, binary = cv.threshold(gray, 0, 255, cv.THRESH_BINARY | cv.THRESH_OTSU)
    binary[binary == 255] = 1
    skel, distance = morphology.medial_axis(binary, return_distance=True)
    dist_on_skel = distance * skel
    skel_img = dist_on_skel.astype(np.uint8)*255
    contours, hireachy = cv.findContours(skel_img, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
    cv.drawContours(image, contours, -1, (0, 0, 255), 1, 8)

    cv.imshow("result", image)
    cv.waitKey(0)
    cv.destroyAllWindows()

03

使用OpenCV实现骨架提取

OpenCV的图像细化的骨架提取方法在扩展模块中,因此需要直接安装opencv-python的扩展包。此外还可以通过形态学的膨胀与腐蚀来实现二值图像的骨架提取,下面的代码实现就是分别演示了基于OpenCV的两种骨架提取方法。代码分别如下:

代码语言:javascript复制
def morph_find(image):
    gray = cv.cvtColor(image, cv.COLOR_BGR2GRAY)
    ret, binary = cv.threshold(gray, 0, 255, cv.THRESH_BINARY | cv.THRESH_OTSU)
    kernel = cv.getStructuringElement(cv.MORPH_CROSS, (3, 3))
    finished = False
    size = np.size(binary)
    skeleton = np.zeros(binary.shape, np.uint8)
    while (not finished):
        eroded = cv.erode(binary, kernel)
        temp = cv.dilate(eroded, kernel)
        temp = cv.subtract(binary, temp)
        skeleton = cv.bitwise_or(skeleton, temp)
        binary = eroded.copy()

        zeros = size - cv.countNonZero(binary)
        if zeros == size:
            finished = True

    contours, hireachy = cv.findContours(skeleton, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
    cv.drawContours(image, contours, -1, (0, 0, 255), 1, 8)
    cv.imshow("skeleton", image)
    cv.waitKey(0)
    cv.destroyAllWindows()


def thin_demo(image):
    gray = cv.cvtColor(image, cv.COLOR_BGR2GRAY)
    ret, binary = cv.threshold(gray, 0, 255, cv.THRESH_BINARY | cv.THRESH_OTSU)
    thinned = cv.ximgproc.thinning(binary)
    contours, hireachy = cv.findContours(thinned, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
    cv.drawContours(image, contours, -1, (0, 0, 255), 1, 8)
    cv.imshow("thin", image)
    cv.waitKey(0)
    cv.destroyAllWindows()

运行结果如下:

0 人点赞