OpenCV图像分割之分水岭算法与图像金字塔算法详解

dst = cv2.distanceTransform(src, distanceType, maskSize[, dstType])

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

# cv2.distanceTransform()函数，计算非0值像素点到0值(背景)像素点的距离
img = cv2.imread('home.jpg')
cv2.imshow('original', img)

img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) 										# 灰度操作
cv2.imshow('gray', img_gray)

ret, thresh = cv2.threshold(img_gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU) 	# 阈值操作
cv2.imshow('thresh', thresh)

kernel = np.ones((3, 3), np.uint8)
img_open = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=2) 				# 形态学操作
cv2.imshow('open', img_open)

img_dilate = cv2.dilate(img_open, kernel, iterations=2)    								# 膨胀操作，确定背景
cv2.imshow('img_dilate', img_dilate)

img_dist = cv2.distanceTransform(img_dilate , cv2.DIST_L2, 5) 							# 距离转换
cv2.imshow('img_dist', img_dist)

cv2.waitKey(0)
cv2.destroyAllWindows()

ret, labels = cv2.connectedComponents(image[, connectivity[, ltype]])

# cv2.connectedComponents()函数，将图像中的背景标记为0，将其他图像标记为从1开始的整数
ret, imgfg = cv2.threshold(img_dist, 0.7*img_dist.max(), 255, cv2.THRESH_TRUNC)

imgfg = np.uint8(imgfg)

ret, markers = cv2.connectedComponents(imgfg)

plt.imshow(imgfg,cmap='gray')
plt.title('imgfg') 
plt.axis('off')
plt.show()

plt.imshow(markers,cmap='gray')
plt.title('markers') 
plt.axis('off')
plt.show()

ret = cv2.watershed(image, markers)

# cv2.watershed()函数，执行分水岭算法分割图像
img = cv2.imread('qizi.jpg')

img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) # 灰度操作

ret, thresh = cv2.threshold(img_gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU) 	# 阈值操作

kernel = np.ones((3, 3), np.uint8)
img_open = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=2) 				# 形态学操作

img_dilate = cv2.dilate(img_open, kernel, iterations=3)               					# 膨胀操作(确定背景)

img_dist = cv2.distanceTransform(img_open, cv2.DIST_L2, 0)          

ret, img_fg = cv2.threshold(img_dist, 0.7*img_dist.max(), 255, 2)      					# 距离操作(确定前景)

img_fg = np.uint8(img_fg)

ret, markers = cv2.connectedComponents(img_fg)

unknown = cv2.subtract(img_dilate, img_fg)												# 确定位置未知区域

markers = markers + 1																	# 加1使背景不为0

markers[unknown == 255] = 0																# 将未知区域设置为0

img_water = cv2.watershed(img, markers)													# 执行分水岭算法

plt.imshow(img_water, cmap='gray')
plt.title('watershed')
plt.axis('off')
plt.show()

img[img_water == -1] = [0, 255, 0]														# 将原图中的被标记点设置为绿色
cv2.namedWindow('watershed', cv2.WINDOW_NORMAL)
cv2.imshow('watershed', img)

cv2.waitKey(0)
cv2.destroyAllWindows()

ret = cv2.pyrDown(image[, dstsize[, borderType]])

# 高斯金字塔向下采样
img = cv2.imread('qizi.jpg')
img1 = cv2.pyrDown(img)
img2 = cv2.pyrDown(img1)

cv2.imshow('img', img)
cv2.imshow('img1', img1)
cv2.imshow('img2', img2)

print('0层的形状：', img.shape)
print('1层的形状：', img1.shape)
print('2层的形状：', img2.shape)

cv2.waitKey(0)
cv2.destroyAllWindows()

ret = cv2.pyrUp(image[, dstsize[, borderType]])

# 拉普拉斯金字塔
img = cv2.imread('qizi.jpg')
img1 = cv2.pyrDown(img)
img2 = cv2.pyrDown(img1)
img3 = cv2.pyrDown(img2)

imgL0 = cv2.subtract(img, cv2.pyrUp(img1))
imgL1 = cv2.subtract(img1, cv2.pyrUp(img2))
imgL2 = cv2.subtract(img2, cv2.pyrUp(img3))

cv2.imshow('imgL0', imgL0)
cv2.imshow('imgL1', imgL1)
cv2.imshow('imgL2', imgL2)

cv2.waitKey(0)
cv2.destroyAllWindows()

# 应用图像金字塔实现图像融合
img1 = cv2.imread('jiang1.jpg')
img2 = cv2.imread('jiang2.jpg')

#生成图像1的高斯金字塔，向下采样6次
img = img1.copy()
img1Gaus = [img]
for i in range(6):
    img = cv2.pyrDown(img)
    img1Gaus.append(img)

#生成图像2的高斯金字塔，向下采样6次
img = img2.copy()
img2Gaus = [img]
for i in range(6):
    img = cv2.pyrDown(img)
    img2Gaus.append(img)

#生成图像1的拉普拉斯金字塔，6层
img1Laps = [img1Gaus[5]]
for i in range(5,0,-1):
    img = cv2.pyrUp(img1Gaus[i])
    lap = cv2.subtract(img1Gaus[i-1],img)    #两个图像大小不同时，做减法会出错
    img1Laps.append(lap)

#生成图像2的拉普拉斯金字塔，6层
img2Laps = [img2Gaus[5]]
for i in range(5,0,-1):
    img = cv2.pyrUp(img2Gaus[i])
    lap = cv2.subtract(img2Gaus[i-1],img)
    img2Laps.append(lap)

#拉普拉斯金字塔拼接：图像1每层左半部分与和图像2每层右半部分拼接
imgLaps = []
for la,lb in zip(img1Laps,img2Laps):
    rows,cols,dpt = la.shape
    ls=la.copy()
    ls[:,int(cols/2):]=lb[:,int(cols/2):]
    imgLaps.append(ls)

#从拉普拉斯金字塔恢复图像
img = imgLaps[0]
for i in range(1,6):
    img = cv2.pyrUp(img)
    img = cv2.add(img, imgLaps[i])

#图像1原图像的半部分与和图像2原图像的右左半部分直接拼接
direct = img1.copy()
direct[:,int(cols/2):]=img2[:,int(cols/2):]
cv2.imshow('Direct',direct)             #显示直接拼接结果
cv2.imshow('Pyramid',img)               #显示图像金字塔拼接结果
cv2.waitKey(0)
cv2.destroyAllWindows()