Morphology of image

1 Probability of morphology

   Image morphology refers to the morphological features of processed images , In essence, convolution kernel is used to calculate the shape or feature of the extracted image .
   Morphological processing is generally black-and-white pixel pictures （ The pixel is 0 perhaps 255）
Common morphological processing methods ：

• Expansion and corrosion
• Open operation
• Closed operation
• Top hat
• Black hat

2 Image global binarization

    Global binarization ： Turn each pixel into 0or255
   cv2.threshold(src, thresh, maxval, type)：
      src: picture , Preferably a grayscale image
     thresh： threshold , How about being greater than a certain value , How about less than a certain value
      maxval： Maximum , Is not necessarily 255
      type： Operation type ：
        cv2.THRESH_BINARY ： Exceeding the threshold is max
        cv2.THRESH_BINARY_INV： Exceeding the threshold is 0
        cv2.THRESH_TOZERO： When the threshold is exceeded, it remains unchanged , Less than change 0

import cv2
import numpy as np

img = cv2.imread('imgs/66.jpg')

#  Binarization of gray-scale photos , Grayscale first 
gary = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

#  Two valued , Return two values ： threshold + Processed pictures 
#  The higher the threshold, the darker , Reasonable adjustment 
thresh, dst = cv2.threshold(gary, 120, 255, cv2.THRESH_BINARY)

cv2.imshow(' Two valued ', np.hstack((gary, dst)))
cv2.waitKey(0)

3 Adaptive threshold binarization

   Using global binarization can not get a suitable threshold for some pixels of the image , Through the automatic calculation of a certain pixel of the picture , Get the corresponding threshold .

• Parameter description ：
     
  cv2.adaptiveThreshold(src, maxValue, adaptiveMethod, thresholdType, blockSize, C)
• adaptiveMethod Method of calculating threshold ：
      -cv2.ADAPTIVE_THRESH_MEAN_C： Take the average value of adjacent areas
      -cv2.ADAPTIVE_THRESH_GAUSSIAN_C： A weighted sum of adjacent areas , Gaussian window
• blockSize： The adjacent threshold size when calculating the threshold
• C： constant ： Average or Subtract this constant from the weighted average

import cv2
import numpy as np

img = cv2.imread('imgs/66.jpg')

#  Binarization of gray-scale photos , Grayscale first 
gary = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

#  Adaptive binarization 
dst = cv2.adaptiveThreshold(gary,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY,9,0)

cv2.imshow(' Adaptive binarization ', np.hstack((gary, dst)))
cv2.waitKey(0)

4 Corrosion operation

    After convolution kernel scanning of corrosion operation , It is white only when all the contents covered are white , Otherwise it's black .

               $K$ = $\left[\begin{array}{ccc}1& 1& 1\\ 1& 1& 1\\ 1& 1& 1\end{array}\right]$

Parameter description ：

• cv2.erode(src,kernel)

import cv2
import numpy as np

img = cv2.imread('imgs/66.jpg')

#  Binarization of gray-scale photos , Grayscale first 
gary = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

#  Corrosion operation 
kernel = np.zeros((3, 3), np.uint8)
dst = cv2.erode(gary, kernel)

cv2.imshow(' Corrosion operation ', np.hstack((gary, dst)))
cv2.waitKey(0)

5 Morphological convolution kernel

   The above method also needs to be called manually kernel, Use opencv Built in api Convolution kernel can be obtained directly
  cv2.getStructuringElement(shape,ksize)

• shape Is the size of the convolution kernel ,33,55 etc.
• cv2.MORPH_RECT : The value of convolution kernel is 1 Matrix
• cv2.MORPH_ELLIPSE: The ellipse
• cv2.MORPH_CROSS: cross

import cv2
import numpy as np

img = cv2.imread('imgs/66.jpg')
img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

# Get the convolution kernel 
kernel = cv2.getStructuringElement(cv2.MORPH_RECT,(5,5))

# Corrosion operation 
std = cv2.erode(img,kernel)
cv2.imshow(' Automatically obtain convolution kernel ',np.hstack((img,std)))

cv2.waitKey(0)

6 Expansion operation

    The expansion operation is completely opposite to the corrosion operation ,
      Enlarge the pixel by one circle , Turn white

• cv2.dilate(scr,kernel)

import cv2
import numpy as np

img = cv2.imread('imgs/66.jpg')
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

#  Generate convolution kernel 
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))

#  Expansion operation 
std = cv2.dilate(img, kernel)

#  display picture 
cv2.imshow(' inflation ', np.hstack((img, std)))

cv2.waitKey(0)

7 Open operation

    Open operation = corrosion + inflation

• cv2.morphologyEx(img, cv2.MORPH_OPEN, kernel,iterations=2)

import cv2
import numpy as np

img = cv2.imread('imgs/66.jpg')
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

#  Generate convolution kernel 
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))

#  Use the open operation ： Corrosion before expansion 
std = cv2.morphologyEx(img, cv2.MORPH_OPEN, kernel,iterations=2)

#  display picture 
cv2.imshow(' inflation ', np.hstack((img, std)))

cv2.waitKey(0)

8 Closed operation

    Expand first and then corrode
  cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel,iterations=2)
    

import cv2
import numpy as np

img = cv2.imread('imgs/66.jpg')
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

#  Generate convolution kernel 
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))

#  Use closed operations ： Expand first and then corrode . Used to remove the noise of content 
std = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel,iterations=2)

#  display picture 
cv2.imshow(' inflation ', np.hstack((img, std)))

cv2.waitKey(0)

9 Morphological gradients

• Basic concepts ： The original image minus the image after corrosion ： That is, the picture edge
• std = cv2.morphologyEx(img, cv2.MORPH_GRADIENT, kernel,iterations=1)

import cv2
import numpy as np

img = cv2.imread('imgs/66.jpg')
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

#  Generate convolution kernel 
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))

#  Get the edge of morphology 
std = cv2.morphologyEx(img, cv2.MORPH_GRADIENT, kernel,iterations=1)

#  display picture 
cv2.imshow(' inflation ', np.hstack((img, std)))

cv2.waitKey(0)

10 Top hat operation

    Top hat = Original picture - Open operation
    Open operation can pick out the noise outside the picture
    namely ： Top hat operation can get rid of the noise

• std = cv2.morphologyEx(img, cv2.MORPH_TOPHAT, kernel, iterations=1)

import cv2
import numpy as np

img = cv2.imread('imgs/66.jpg')
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

#  Generate convolution kernel 
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))

#  Top hat operation , Get the noise outside the picture 
std = cv2.morphologyEx(img, cv2.MORPH_TOPHAT, kernel, iterations=1)

#  display picture 
cv2.imshow(' inflation ', np.hstack((img, std)))

cv2.waitKey(0)

11 Black hat operation

• Black hat = Original picture - Closed operation
• Close operation is to remove the noise inside the picture
• Black hat gets the internal noise of the picture
• std = cv2.morphologyEx(img, cv2.MORPH_BLACKHAT, kernel, iterations=1)

import cv2
import numpy as np

img = cv2.imread('imgs/66.jpg')
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

#  Generate convolution kernel 
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))

#  Black hat operation , Get the noise in the picture 
std = cv2.morphologyEx(img, cv2.MORPH_BLACKHAT, kernel, iterations=1)

#  display picture 
cv2.imshow(' inflation ', np.hstack((img, std)))

cv2.waitKey(0)