此書極好，值得借鑒學習，並且開源開放。Python在實現過程中，體現出來了非常強的優勢，特別是結合Numpy來進行矩陣計算，有很多簡化方法。這里將學習過程代碼進行增編、添加后進行展示。
Python目前的缺點應該是缺乏一個像ImageWatch這樣的工具，這將影響算法研究；另外Numpy的過度抽象，某種程度上也會造成障礙。
1、尋找指定色彩區域
Python的特色，在於Numpy的使用
import cv2
import numpy
as np
src = cv2.imread(
"e:/template/tiantan.jpg")
hsv = cv2.cvtColor(src,cv2.COLOR_BGR2HSV)
lower_blue = np.array([
100,
43,
46])
upper_blue = np.array([
124,
255,
255])
mask = cv2.inRange(hsv,lower_blue,upper_blue)
res = cv2.bitwise_and(src,src,mask=mask)
cv2.imshow(
"hsv",hsv)
cv2.imshow(
"mask",mask)
cv2.imshow(
"res",res)
cv2.waitKey(
0)
2、warpperspective 透視變化的python實現
import cv2
import numpy
as np
src = cv2.imread(
"e:/template/steel03.jpg")
rows,cols,ch = src.shape
pts1 = np.float32([[
122,
0],[
814,
0],[
22,
540],[
910,
540]])
pts2 = np.float32([[
0,
0],[
960,
0],[
0,
540],[
960,
540]])
M = cv2.getPerspectiveTransform(pts1,pts2)
dst = cv2.warpPerspective(src,M,(cols,rows))
cv2.imshow(
"src",dst)
cv2.waitKey(
0)
這里操作的核心，是一個np的矩陣。在C++中，使用Vector，可能會造成很多浪費。
3、自適應閾值
import cv2
import numpy
as np
obj = cv2.imread(
"e:/template/pig.jpg",
0)
ret,th1 = cv2.threshold(obj,
100,
255,cv2.THRESH_BINARY)
th2 = cv2.adaptiveThreshold(obj,
255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY,
11,
2)
ret3,th3 = cv2.threshold(obj,
0,
255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)
cv2.imshow(
"th3",th3)
print(ret3)
cv2.waitKey()
當參數選擇OTSU的時候，能夠根據計算，自動算出下限。但是我認為這一點並沒有什么特別的用途。
4、模糊處理
obj = cv2.imread(
"e:/template/pig.jpg",
0)
blur= cv2.blur(obj,(
3,
3))
gaussBlur=cv2.GaussianBlur(obj,(
3,
3),
0)
median = cv2.medianBlur(obj,
5)
bilate = cv2.bilateralFilter(obj,
0.75,
0.75)
5、形態學變換
obj = cv2.imread(
"e:/template/pig.jpg",
0)
opening = cv2.morphologyEx(obj,cv2.MORPH_OPEN,(
7,
7))
cv2.imshow(
"obj",obj)
cv2.imshow(
"opening",opening)
我喜歡這種寫法，這將有長遠影響。
6、梯度變化，包括1階、2階和混合的。
obj = cv2.imread(
"e:/template/pig.jpg",
0)
laplacian = cv2.Laplacian(obj,cv2.CV_64F)
sobelx=cv2.Sobel(obj,cv2.CV_64F,
1,
0,ksize=
5)
sobely=cv2.Sobel(obj,cv2.CV_64F,
0,
1,ksize=
5)
7、梯度融合
曾經這段代碼很神秘的，但是今日使用Python來寫，就非常簡單。可以看出，Python用來處理二維矩陣信息是很強的。
# Standard imports
import cv2
import numpy
as np
A = cv2.imread(
"e:/template/apple.jpg")
B = cv2.imread(
"e:/template/orange.jpg")
G = A.copy()
gpA=[G]
for i
in range(
6):
G = cv2.pyrDown(G)
gpA.append(G)
G = B.copy()
gpB = [G]
for i
in range(
6):
G = cv2.pyrDown(G)
gpB.append(G)
lpA = [gpA[
5]]
for i
in range(
5,
0,-
1):
GE = cv2.pyrUp(gpA[i])
L = cv2.subtract(gpA[i-
1],GE)
lpA.append(L)
lpB = [gpB[
5]]
for i
in range(
5,
0,-
1):
GE = cv2.pyrUp(gpB[i])
L = cv2.subtract(gpB[i-
1],GE)
lpB.append(L)
LS = []
for la,lb
in zip(lpA,lpB):
rows,cols,dpt= la.shape
print(rows,cols)
ls = np.hstack((la[:,
0:cols//
2],lb[:,cols//
2:]))
#直接橫向排列
LS.append(ls)
ls_ = LS[
0]
for i
in range(
1,
6):
ls_ = cv2.pyrUp(ls_)
ls_ = cv2.add(ls_,LS[i])
real = np.hstack((A[:,:cols//
2],B[:,cols//
2:]))
cv2.imshow(
"ls_",ls_)
cv2.imshow(
"real",real)
cv2.waitKey()
8、輪廓尋找
import cv2
import numpy
as np
src = cv2.imread(
"e:/template/rectangle.jpg")
gray = cv2.cvtColor(src,cv2.COLOR_BGR2GRAY)
ret,thresh = cv2.threshold(gray,
127,
255,
0)
contours,hierarchy = cv2.findContours(thresh,cv2.RETR_TREE,cv2.CHAIN_APPROX_NONE)
print(contours)
src = cv2.drawContours(src,contours,-
1,(
0,
255,
0),
3)
cv2.imshow(
"src",src)
cv2.waitKey()
這里，使用
cv2.CHAIN_APPROX_NONE
或者不同的參數的話，會獲得不同的輪廓結果。這對於我現有的輪廓分析研究，也是有幫助的。
9、輪廓的最小 接圓和最大內切圓
外接圓比較簡單
(x,y),radius = cv2.minEnclosingCircle(contours[
0])
center = (int(x),int(y))
radius = int(radius)
src = cv2.circle(src,center,radius,(
0,
255,
0),
2)
注意它這里的表示方法。內切圓則采用特殊方法。
# Get the contours
contours, _ = cv.findContours(thresh, cv.RETR_TREE, cv.CHAIN_APPROX_SIMPLE)
# Calculate the distances to the contour
raw_dist = np.empty(thresh.shape, dtype=np.float32)
for i
in range(src.shape[
0]):
for j
in range(src.shape[
1]):
raw_dist[i,j] = cv.pointPolygonTest(contours[
0], (j,i),
True)
minVal, maxVal, _, maxDistPt = cv.minMaxLoc(raw_dist)
minVal = abs(minVal)
maxVal = abs(maxVal)
# Depicting the distances graphically
drawing = np.zeros((src.shape[
0], src.shape[
1],
3), dtype=np.uint8)
for i
in range(src.shape[
0]):
for j
in range(src.shape[
1]):
if raw_dist[i,j] <
0:
drawing[i,j,
0] =
255 - abs(raw_dist[i,j]) *
255 / minVal
elif raw_dist[i,j] >
0:
drawing[i,j,
2] =
255 - raw_dist[i,j] *
255 / maxVal
else:
drawing[i,j,
0] =
255
drawing[i,j,
1] =
255
drawing[i,j,
2] =
255
cv.circle(drawing,maxDistPt,int(maxVal),(
255,
255,
255))
cv.imshow(
'Source', src)
cv.imshow(
'Distance and inscribed circle', drawing)
cv.waitKey()
最大內接圓則復雜許多。
10、尋找輪廓的極點
contours, _ = cv2.findContours(thresh,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
cnt = contours[
0]
leftmost = tuple(cnt[cnt[:,:,
0].argmin()][
0])
rightmost= tuple(cnt[cnt[:,:,
0].argmax()][
0])
topmost = tuple(cnt[cnt[:,:,
1].argmin()][
0])
bottommost=tuple(cnt[cnt[:,:,
1].argmax()][
0])
cv2.circle(src,leftmost,
5,(
0,
255,
0))
cv2.circle(src,rightmost,
5,(
0,
255,
255))
cv2.circle(src,topmost,
5,(
255,
255,
0))
cv2.circle(src,bottommost,
5,(
255,
0,
0))
cv2.imshow(
"src",src)
這是一種很好的方法，能夠直接找出輪廓的各方向邊界。
11 模板匹配
src = cv.imread(
"e:/template/lena.jpg",
0)
template = cv.imread(
"e:/template/lenaface.jpg",
0)
w,h = template.shape
res = cv.matchTemplate(src,template,cv.TM_CCOEFF)
min_val,max_val,min_loc,max_loc = cv.minMaxLoc(res)
cv.rectangle(src,max_loc,(max_loc[
0]+w,max_loc[
1]+h),(
0,
0,
255),
2)
cv.imshow(
"template",template)
cv.imshow(
"src",src)
cv.waitKey()
我想體現的是python它的寫法有很大不同。
src = cv.imread(
"e:/template/coin.jpg")
gray = cv.cvtColor(src,cv.COLOR_BGR2GRAY)
template = cv.imread(
"e:/template/coincut.jpg",
0)
w,h = template.shape
res = cv.matchTemplate(gray,template,cv.TM_CCOEFF_NORMED)
threshold =
0.4
loc = np.where(res>=threshold)
print(loc)
for pt
in zip(*loc[::
1]):
cv.rectangle(src,pt,(pt[
0]+w,pt[
1]+h),(
0,
0,
255),
2)
cv.imshow(
"template",template)
cv.imshow(
"src",src)
cv.waitKey()
結合使用閾值，可以實現多目標匹配。
# Standard imports
import cv2
as cv
import numpy
as np
src = cv.imread(
"e:/template/coin.jpg")
gray = cv.cvtColor(src,cv.COLOR_BGR2GRAY)
template = cv.imread(
"e:/template/coincut.jpg",
0)
w,h = template.shape
res = cv.matchTemplate(gray,template,cv.TM_CCOEFF_NORMED)
threshold =
0.6
loc = np.where(res>=threshold)
print(loc)
for pt
in zip(*loc[::-
1]):
#排序方法為height width
print(pt)
cv.rectangle(src,pt,(pt[
0]+w,pt[
1]+h),(
0,
0,
255),
2)
cv.imshow(
"template",template)
cv.imshow(
"src",src)
cv.waitKey()
特別需要注意其排序方法。但是這里的閾值選擇，也是超參數類型的。
12 HoughCircle
src = cv.imread(
"e:/template/circle.jpg",
0)
src = cv.medianBlur(src,
5)
cimg = cv.cvtColor(src,cv.COLOR_GRAY2BGR)
circles = cv.HoughCircles(src,cv.HOUGH_GRADIENT,
1,
20,param1=
50,param2=
30,minRadius=
0,maxRadius=
0)
circles = np.uint16(np.around(circles))
for i
in circles[
0,:]:
cv.circle(cimg,(i[
0],i[
1]),i[
2],(
0,
255,
0),
2)
cv.circle(cimg,(i[
0],i[
1]),
2,(
0,
0,
255),
3)
cv.imshow(
"src",cimg)
cv.waitKey()
13 風水嶺算法
# Standard imports
import cv2
as cv
import numpy
as np
src = cv.imread(
"e:/template/water_coins.jpg")
gray =cv.cvtColor(src,cv.COLOR_BGR2GRAY)
_,thresh = cv.threshold(gray,
0,
255,cv.THRESH_BINARY_INV+cv.THRESH_OTSU)
kernel = np.ones((
3,
3),np.uint8)
opening = cv.morphologyEx(thresh,cv.MORPH_OPEN,kernel,iterations=
2)
sur_bg = cv.dilate(opening,kernel)
dist_transform = cv.distanceTransform(opening,
1,
5)
_,sur_fg=cv.threshold(dist_transform,
0.7*dist_transform.max(),
255,
0)
sur_fg = np.uint8(sur_fg)
unknow = cv.subtract(sur_bg,sur_fg)
_,markers1 = cv.connectedComponents(sur_fg)
markers = markers1+
1
markers[unknow ==
255] =
0
markers3 = cv.watershed(src,markers)
src[markers3 == -
1] = [
255,
0,
0]
cv.imshow(
"src",src)
cv.waitKey()
這個結果，具有參考價值。