[opencv learning] [Canny edge detection]

Let's learn today Canny testing , The specific algorithm is shown in the code .

import cv2
import numpy as np

#  edge detection , The characteristic of edge points is that the gradient is large , General non edge points , The gradient is smaller or even 0
# Canny  Edge detection process ：
# 1： Use Gaussian filter to smooth the image , Remove noise points 
# 2： Calculate the gradient and direction of each pixel in the image , This method can be used Sobel operator X and Y Direction get 
# 3： Apply non maximum suppression （Non-Maximum-suppression）, Eliminate some edge effects of edge detection , Because it is likely that adjacent pixels may be considered edge points , At this time, we have to remove some unnecessary interference points , Take the most likely edge pixels 
# 4： Apply double threshold method , Identify real potential boundaries , Remove some unnecessary noise points 
# 5： Then suppress the isolated edge （ It is considered as noise or non critical edge points ） Complete the purpose 

#  Disassembly instructions 
# 1:  Gauss filter , We are familiar with this , Use the following kernel for convolution 
#  such as ：
# H = [[0.0924, 0.1192, 0.0924],
# [0.1192, 0.1538, 0.1192],
# [0.0924, 0.1192, 0.0924]]
# img = img * H

# 2： Calculate the gradient and direction of pixels 
#  Use here Sobel operator 
# Sx = [[-1, 0, 1], [-2, 0, 2], [-1, 0, 1]]
# Sy = [[-1, -2, -1], [0, 0, 0], [1, 2, 1]]
#  Convolute with the image machine respectively , Get the horizontal gradient Gx, And vertical gradient Gy
# Gx = img * Sx
# Gy = img * Sy
#  utilize  Gx  and  Gy  Calculate the total gradient intensity G  And direction  θ
# G = (Gx^2 + Gy^2)^0.5
# θ = arctan(Gy / Gx), This is to determine which two adjacent pixels the central pixel is compared with 

# 3： Non maximum suppression 
#  For simplicity of calculation , There are eight directions around a pixel , The angles are (0, 45,90, 135, 180, 225, 270, 315)
# θ And which angle range is closest , It belongs to which angle , In each angle direction, you can find the center point near N A numerical （ Gradient intensity value ）,N Depending on kernel size Size 
#  The gradient intensity value of the center point and   Near N Compare the gradient intensity values of points , If the gradient intensity value of the center point is the largest , Then keep the strength value , Otherwise, it will be suppressed , Change to 0.

# 4： Double threshold method 
#  Set a maximum gradient intensity  max,  And gradient intensity minimum  min
#  Filter the strength calculated above ,
#  The gradient intensity value is less than  min Of , Direct suppression , Change it to 0.
#  The gradient intensity value is greater than  max Of , Keep... Directly .
#  Gradient intensity is between the two , See if this value is connected with other boundaries , If there is a link, keep it , Otherwise, the suppression is 0.

img = cv2.imread('images/saoge.jpg', cv2.IMREAD_GRAYSCALE)
print(img.shape)

can1 = cv2.Canny(img, threshold1=20, threshold2=100)
print(can1.shape)
can2 = cv2.Canny(img, threshold1=80, threshold2=160)
print(can2.shape)

ret = np.hstack((img, can1, can2))
cv2.imshow('Canny', ret)
cv2.waitKey(0)
cv2.destroyAllWindows()

The effect is as follows ：