如何从蒙版绘制正确形状的多边形? [英] How can I draw a polygon in the right shape from a mask?

问题描述

• 我有一个带有不同标记区域的蒙版，如所附图像所示.
• 使用活动轮廓功能进行分割之后，我想在活动轮廓点的合计点周围绘制一个多边形，但是该多边形逐行读取这些点，因此该多边形绘制不正确，如附图所示，则多边形应看起来更像一个圆形区域.

• 我该如何解决?

• I have a mask with different labelled areas as seen in the attached image.
• After doing segmentation with active contour function, I want to draw a polygon around the resultant points of the active contour points, but the polygon reads the points row by row, so the polygon drawing is not correct as seen in the attached figure, the polygon should look like more like a circular area.

• How can I solve it please?

图片

import numpy as np
from skimage.io import imread, imsave
import skimage.filters as filters
from skimage.segmentation import active_contour
from scipy.sparse import csr_matrix
import matplotlib.pyplot as plt
from skimage.draw import polygon
#---------------------------------
img  = imread('image.jpg')
mask = imread('label.png')

def activeContour(img, mask, alpha=0.015, beta=10, gamma=0.001, iterations=500, max_step=0.5):
    lb_list = np.unique(mask)
    Result = np.zeros_like(img)
    for lb in lb_list:
        if(lb != 0):
            s = np.where(mask == lb)
            r = s[0].astype(np.uint8)
            c = s[1].astype(np.uint8)
            init = np.array([r, c]).T
            #-------PreProcessing----------------------------
            img2 = filters.gaussian(img, 3)
            #------- Active Contour  ------------------------
            snake = active_contour(img2, init, alpha=alpha, beta=beta, gamma=gamma, coordinates='rc',
             max_px_move=max_step, max_iterations=iterations)
            #------------------------------------------------
            vals = lb*np.ones_like(snake[:, 0],dtype=np.uint8)
            label = csr_matrix((vals, (snake[:, 1].astype(np.uint8), snake[:, 0].astype(np.uint8))), shape=(img.shape[0], img.shape[1])).toarray()
            Result = Result + label
    return Result


res = activeContour(img, mask, alpha=0, beta= 10, gamma= 0.0000001)
labels = np.unique(res)
ctr = 25
for label in labels:
    if label != 0:
        img = np.zeros_like(img)
        s = np.where(res == label)
        r = s[0].astype(np.uint8)
        c = s[1].astype(np.uint8)
        rr, cc = polygon(r, c)
        img[rr, cc] = 2*ctr
        plt.imshow(img, cmap='gray')
        plt.title('%d'%label)
        plt.axis('off')
        plt.show()
        ctr+=1

推荐答案

感谢 Juan 对他的帮助，理论上此答案的一部分是此处并且实施失败了.

Thanks for Juan for his help, the theoritical part of this answer is here and the implementation is down.

import numpy as np

def sortVertices(array):
    s = np.where(array != 0)
    yr = s[0].astype(np.uint8)
    xc = s[1].astype(np.uint8)
    center_xc = np.sum(xc)/xc.shape 
    center_yr = np.sum(yr)/yr.shape 
    theta = np.arctan2(yr-center_yr, xc-center_xc) * 180 / np.pi
    indices = np.argsort(theta)
    x = xc[indices]
    y = yr[indices]
    return x, y

这篇关于如何从蒙版绘制正确形状的多边形?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！