Because recently, company projects need to use opencv Process the image , So I learned this piece , Now record what you have learned , For use .

One 、 Image reading and display

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
cv2.imshow("soble",img)
cv2.waitKey(10000)

But there's a hole , If it is a Chinese path , You will find that the picture cannot be read , Need a different way

#path Path for picture 
img_np = np.fromfile(path, dtype = np.uint8)   
imgs = cv2.imdecode(img_np, -1) 

Two 、 Basis of image processing

1. Pixel processing

Read pixel

When reading picture pixels , If the image is a grayscale image, the grayscale value is returned , If it is BGR Images , The return value is B,G,R Value .

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
# If it is a grayscale image, there is only one return value , If BGR The image returns three values 
p = img[11,11]
print(p)

  Normally, the color picture returns BGR The value of three channels , If you want to get a channel value separately, this is in , Add a passage after the position

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
# If it is a grayscale image, there is only one return value , If BGR The image returns three values , If the channel value is added, only the value of the current channel will be returned 【0,1,2】 It represents three channels of blue, green and red 
blue = img[11,11,0]
# At this time, the blue channel value is returned 
print(blue)

green = img[11,11,1]
# At this time, the green channel value is returned 
print(green)


red = img[11,11,2]
# At this time, the red channel value is returned 
print(red)

Change the pixel value

  If it is a grayscale image, the modification method is ：

img[11,11] = 255

If BGR The image is knee style ：

img[11,11] = [255,255,255]

perhaps

img[11,11,0] = 255

img[11,11,1] = 255

img[11,11,2] = 255

2. Use numpy visit

Read pixel  

  This operation is basically the same as the above . It is also divided into grayscale pictures and BGR Two kinds of pictures , The return value is the same . If it is a grayscale image, only one value is returned , If it is BGR The picture returns three values

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
blue = img.item(78,125,0)
green = img.item(78,125,1)
red = img.item(78,125,2)
print(blue)
print(green)
print(red)

Change the pixel value

  Compared to not using numpy,numpy The setting of pixels is more convenient .

The operation mode is as follows ：

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
# Grayscale image ： Receive two parameters , The first parameter is pixel position , The second position is the pixel value 
img.itemset((88,99),255)


#BGR Images    It also receives two parameters , The first parameter is pixel position , But there is one more channel , The second is the pixel value 
img.itemset((88,99,0),255)
img.itemset((88,99,0),255)
img.itemset((88,99,0),255)

3. Get image properties

opencv The attributes of image acquisition include three aspects

• shape ： That's ok 、 Column 、 The channel number
• Number of pixels
• The data type of the image

1. shape

shape You can get the shape of the image

Returns the number of containing rows , Number of columns , Tuple of the number of channels . If it is a grayscale image, return the number of rows and columns , If it is BGR The image returns the number of lines 、 Number of columns 、 The channel number

import cv2
import numpy as np

# Read images 
img = cv2.imread("soble.png")
# Turn the image into a grayscale image 
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

print(img.shape)    # The output value is (428, 605, 3)

print(gray.shape)  # The output value is (428, 605)

 2. Number of pixels

 size The number of pixels of the image can be obtained

If it is a grayscale image, return the number of lines * Number of columns , If it is BGR The image returns the number of lines * Number of columns * Number of channels

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

print(img.size)    # The output value is 776820
print(gray.size)   # Output value 258940

  It can be seen that after turning the picture into a gray image , There is only one channel left in the picture , Compared with BGR The picture pixels are reduced a lot

3. Image type

 dtype The data type of the image is returned

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

print(img.dtype)   #uint8
print(gray.dtype)  #uint8

4. Region of interest （ROI）

ROI(region of interest), Region of interest

From the processed image to the box 、 round 、 The ellipse 、 Irregular polygon and other ways to outline the area to be dealt with .

It can be done by various operators （operator） And function to find the region of interest ROI, And the next step of image processing .

import cv2
import numpy as np
 
img = cv2.imread("soble.png")

cv2.imshow('ori',img)

temp = img[100:300,100:300]
img[100:300,400:600] = temp

cv2.imshow('temp',img)
cv2.waitKey()
cv2.destroyAllWindows()

5. Splitting and merging of channels

1. Split channels

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
cv2.imshow('ori',img)

b = img[:,:,0]
g = img[:,:,1]
r = img[:,:,2]
cv2.imshow('b',b)
cv2.imshow('g',g)
cv2.imshow('r',r)
cv2.waitKey()
cv2.destroyAllWindows()

You can also use opencv The provided tool class performs channel splitting

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
cv2.imshow('ori',img)

b,g,r = cv2.split(img)
cv2.imshow('b',b)
cv2.imshow('g',g)
cv2.imshow('r',r)
cv2.waitKey()
cv2.destroyAllWindows()

  The results of the two splitting methods are the same , It's shown as follows ：

2. Merge channel  

  adopt opencv Provided tool classes for operation

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
cv2.imshow('ori',img)
b,g,r = cv2.split(img)
result = cv2.merge([b,g,r])
cv2.imshow('result',result)
cv2.waitKey()
cv2.destroyAllWindows()

  You can see that the images after merging are consistent with those before splitting .

3、 ... and 、 Image operation

1. Image addition

Numpy Add

numpy Addition adopts modular addition , If two pixels are added , When the pixel value is less than 255 When, the pixel value of the current addition is taken , If the added pixel value is greater than 255 If so, take the current pixel value pair 255 modulus

OpenCV Add

OpenCV Addition adopts saturation operation addition , How it works ： result = cv2.add( Images 1, Images 2)

When adding with saturation operation , If the added pixel value is lower than 255 Then take the current added pixel value , If the added pixel value is greater than 255 Then take 255 As the current pixel value .

  Something to be aware of （ The size of the image involved in the operation 、 The type must be the same ）

The following code is used to operate ：

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
cv2.imshow('ori',img)
a = img + img
b = cv2.add(img,img)
cv2.imshow('a',a)
cv2.imshow('b',b)
cv2.waitKey()
cv2.destroyAllWindows()

It can be seen that there is a big problem in the direct addition of this image , After adding, the picture is very different from the original picture . So the next step is to introduce the commonly used image fusion . Compared with this simple image addition effect will be better .

2. Image fusion

  The resulting image = Images 1* coefficient 1 + Images 2* coefficient 2 + The amount of brightness adjustment

img = img1*0.3 + img2*0.7 +18

Image fusion function addWeighted

dst = cv.addWeighted(src1,alpha,src2,beta,gamma)

dst = src1*alpha + src2*beta + gamma;

It's important to note that the parameters gamma Don't omit

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
cv2.imshow('ori',img)
result = cv2.addWeighted(img,0.3,img,0.7,0)
cv2.imshow('result',result)
cv2.waitKey()
cv2.destroyAllWindows()

  It was found by experiment that , You can find alpha and beta It is to control the display proportion of pictures , Which picture has a large coefficient , Then which picture is more obvious .

Four 、 Type conversion

Convert an image from one type to another

OpenCV Provides 200 Various image conversion methods , If necessary, please refer to the official documents

  A simple example , take BGR turn gray Come on ：

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
cv2.imshow('gray',gray)
cv2.waitKey()
cv2.destroyAllWindows()

5、 ... and 、 Geometric transformation

1. Image zoom

dst = cv2.resize(src,dsize)

disze Zoom size

b = cv2.resize(a,(122,122))

 dst = cv2.resize(src,dsize,fx,fy)

fx,fy Zoom size , If dsize If it is not empty, follow dsize Zoom , If it is empty, follow fx,fy Zoom

b = cv2.resize(a,None,fx=0.5,fy=0.7)

  Code example ：

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
rows,cols = img.shape[:2]
b=cv2.resize(img,(round(cols*0.5),round(rows*1.2)))
cv2.imshow('ori',img)
cv2.imshow('resize',b)
cv2.waitKey()
cv2.destroyAllWindows()

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
rows,cols = img.shape[:2]
b=cv2.resize(img,None,fx=1,fy=0.5)
cv2.imshow('ori',img)
cv2.imshow('resize',b)
cv2.waitKey()
cv2.destroyAllWindows()

2. Image reversal

dst = cv2.flip(src,flipCode)

flipCode = 0 With X Flip the axis of symmetry

flipCode > 0 With y The axis is the symmetry axis to flip

flipCode < 0 With y The axis is the symmetry axis to flip , And then to x The axis is the symmetry axis to flip

  Code example ：

import cv2
import numpy as np
 
img = cv2.imread("soble.png")
b= cv2.flip(img,0)
c= cv2.flip(img,1)
d= cv2.flip(img,-1)
cv2.imshow('ori',img)
cv2.imshow('b',b)
cv2.imshow('c',c)
cv2.imshow('d',d)
cv2.waitKey()
cv2.destroyAllWindows()

Due to time , I'll write here first today, and I'll continue to write the rest of the time .

If you have any questions, you can add wechat communication , Because I have just begun to contact , I hope you can have more exchanges .