Python--level set (水平集)和 chan-vese模型
版权声明:本文为博主原创文章,未经博主允许不得转载。 https://blog.csdn.net/qq_38784098/article/details/82144106
level set :https://www.zhihu.com/question/22608763?sort=created
https://blog.csdn.net/xiangyong58/article/details/11876019
chan-vese模型(公式推导):https://blog.csdn.net/zhangchen1003/article/details/48930377
水平集(CV模型)代码:
-
import cv2
-
from pylab import*
-
-
Image = cv2.imread('02.jpg', 1) # 读入原图
-
image = cv2.cvtColor(Image, cv2.COLOR_BGR2GRAY)
-
img = np.array(image, dtype=np.float64) # 读入到np的array中,并转化浮点类型
-
-
# 初始水平集函数
-
IniLSF = np.ones((img.shape[0], img.shape[1]), img.dtype)
-
IniLSF[300:320, 300:320] = -1
-
IniLSF = -IniLSF
-
-
# 画初始轮廓
-
Image = cv2.cvtColor(Image, cv2.COLOR_BGR2RGB)
-
plt.figure(1), plt.imshow(Image), plt.xticks([]), plt.yticks([]) # to hide tick values on X and Y axis
-
plt.contour(IniLSF, [0], color='b', linewidth=2) # 画LSF=0处的等高线
-
plt.draw(), plt.show(block=False)
-
-
-
def mat_math(intput, str):
-
output = intput
-
for i in range(img.shape[0]):
-
for j in range(img.shape[1]):
-
if str == "atan":
-
output[i, j] = math.atan(intput[i, j])
-
if str == "sqrt":
-
output[i, j] = math.sqrt(intput[i, j])
-
return output
-
-
-
# CV函数
-
def CV(LSF, img, mu, nu, epison, step):
-
-
Drc = (epison / math.pi) / (epison*epison + LSF*LSF)
-
Hea = 0.5*(1 + (2 / math.pi)*mat_math(LSF/epison, "atan"))
-
Iy, Ix = np.gradient(LSF)
-
s = mat_math(Ix*Ix+Iy*Iy, "sqrt")
-
Nx = Ix / (s+0.000001)
-
Ny = Iy / (s+0.000001)
-
Mxx, Nxx = np.gradient(Nx)
-
Nyy, Myy = np.gradient(Ny)
-
cur = Nxx + Nyy
-
Length = nu*Drc*cur
-
-
Lap = cv2.Laplacian(LSF, -1)
-
Penalty = mu*(Lap - cur)
-
-
s1 = Hea*img
-
s2 = (1-Hea)*img
-
s3 = 1-Hea
-
C1 = s1.sum() / Hea.sum()
-
C2 = s2.sum() / s3.sum()
-
CVterm = Drc*(-1 * (img - C1)*(img - C1) + 1 * (img - C2)*(img - C2))
-
-
LSF = LSF + step*(Length + Penalty + CVterm)
-
# plt.imshow(s, cmap ='gray'),plt.show()
-
return LSF
-
-
# 模型参数
-
mu = 1
-
nu = 0.003 * 255 * 255
-
num = 20
-
epison = 1
-
step = 0.1
-
LSF = IniLSF
-
for i in range(1, num):
-
LSF = CV(LSF, img, mu, nu, epison, step) # 迭代
-
if i % 1 == 0: # 显示分割轮廓
-
plt.imshow(Image), plt.xticks([]), plt.yticks([])
-
plt.contour(LSF, [0], colors='r', linewidth=2)
-
plt.draw(), plt.show(block=False), plt.pause(0.01)
为什么上传图片这么麻烦。
一、文章参考
Chan T F, Vese L. Active contours without edges[J]. Image processing, IEEE transactions on, 2001, 10(2): 266-277.
1
二、公式推导过程
---------------------
作者:jonson_zc
来源:CSDN
原文:https://blog.csdn.net/zhangchen1003/article/details/48930377
版权声明:本文为博主原创文章,转载请附上博文链接!