At present, there are several mainstream image databases:

1. OpenCV      2. PIL(Pillow)       3. matplotlib.image     4. skimage      5. scipy.misc

Conclusion: OpenCV is the most powerful and mature image library.

 

1.1 reading and storage of OpenCV image

1.2OpenCV image ashing

import cv2
#Method 1
img = cv2.imread('horse.jpg',0) # 0 Read in gray picture,Or use cv2.IMREAD_GRATSCALE Substitute 0
cv2.imshow('gray image',img)

#Method two
img = cv2.imread('horse.jpg')
gray_img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
cv2.imshow('gray image',gray_img)

print(gray_img.shape) # (height, width)
print(gray_img.size)  # Pixel number
print(gray_img)       # Of the printed image numpy Array, 2D
cv2.waitKey()

1.3 OpenCV matrix format transformation

Why? : OpenCV's matrix format (height, width, channels) - > > the type of deep learning matrix may be (channels, height, width)

import cv2
import numpy as np
img = cv2.imread('horse.jpg',1)
cv2.imshow('image',img)
# Transformation of matrix format
print(img.shape)
img = img.transpose(2,0,1) #Transformation function 
print(img.shape)

# Matrix extension (batch_size, channels, height, width) Operation of forecast single picture
# Add a column as the number of pictures
img = np.expand_dims(img, axis=0) #Use numpy function
print(img.shape)

# Training stage construction batch
data_lst = []
loop:
    img = cv2.imread('xxx.jpg')
    data_lst.append(img)
data_arr = np.array(data_lst)

1.4 OpenCV image normalization

import cv2
# In order to reduce the computation, it is necessary to set the pixel value 0-255 Switch to 0-1 Between
img = cv2.imread('horse.jpg')
img = img.astype('float') / 255.0 # First convert the data type to float
print(img.dtype)
print(img)

1.5 OpenCV BRG to RGB

import cv2
img = cv2.imread('horse.jpg')
img = cv2.cvtColor(img,cv2.COLOR_BGR2RGB) # Turn to RGB format
print(img)

1.6 OpenCV access pixel

import cv2
img = cv2.imread('horse.jpg')
gray_img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY) # Turn to Gray image
print(img[4,4])        # 3 channels
print(gray_img[4,4])   # 1 channel

1.7 OpenCV ROI

import cv2
img = cv2.imread('horse.jpg')
print(img.shape)
roi = img[0:437,0:400] # [y:height,x:width]

cv2.imshow('roi',roi)
cv2.waitKey()

 

 

2.1 PIL image reading and storage

from PIL import Image
import numpy as np
#Image reading
img = Image.open('horse.jpg')
print(img.format) # Picture format
print(img.size)   # (width,height)
print(img.mode)   # Picture channel type

#Convert image to matrix format
arr = np.array(img)
print(arr.shape)
print(arr.dtype)

#Image storage
new_img = Image.fromarray(arr)
new_img.save('test.jpg')

img.show()

2.2 gray processing of PIL image

#Image gray processing
gray = Image.open('horse.jpg').convert('L')
gray_arr = np.array(gray)
print(gray_arr.shape) # (height,width)
print(gray_arr.dtype)
print(gray_arr)
gray.show()

2.3 PIL region of interest shear

# Region of interest cut
img = Image.open('horse.jpg')
roi = img.crop((0,0,200,200)) # (Left upper x，Left upper y，lower right x，lower right y)
roi.show()

2.4 channel operation

# Channel processing
r,g,b = img.split() #separate
img = Image.merge("RGB",(b,g,r)) #merge
img = img.copy() #copy

 

3.1 Matplotlib reading and storing pictures

import matplotlib.pyplot as plt
import numpy as np
# Image read as numpy Array format
img = plt.imread('horse.jpg')

plt.axis('off') # Turn off scale display

print(img.shape) # (height, width, channel)
print(img.size) # Pixel number
print(img.dtype)

#Save pictures
plt.savefig('./name.jpg')

figure = plt.figure(figsize=(20,10)) # Resize display picture

plt.imshow(img)
plt.show()

3.2 image ashing of Matplotlib

#Image ashing
# Average value
img_mean = img.mean(axis=2)
plt.imshow(img_mean,cmap='gray')
plt.show()

#Maximum value method
img_max = img.max(axis=-1)
plt.imshow(img_max,cmap='gray')
plt.show()

#RGB Trichromatic method
gravity = np.array([0.299,0.587,0.114])
img_gravity = np.dot(img,gravity)
plt.imshow(img_gravity,cmap="gray")
plt.show()

 

4.1 skimage reading and storing images

from skimage import io
#Read image numpy Array format
img = io.imread('horse.jpg')
print(img.shape)
print(img.dtype)
print(img.size)
#print(img)
io.imshow(img)

#Save image
io.imsave('test.jpg',img)

4.2 ashing of skimage

#Image ashing and normalization
img = io.imread('horse.jpg',as_gray=True)
print(img.shape)
print(img.dtype) # Data type bit float
print(img.size)
print(img)
io.imshow(img)
io.show()

 

5.1 scipy.misc reading and storing images

#At 1.2.0 After unified use imageio Modular
import imageio
import matplotlib.pyplot as plt
#Read picture as numpy array
img = imageio.imread('horse.jpg')
print(img.dtype)
print(img.size)  # Pixel number
print(img.shape) #(height, width, channels)
plt.imshow(img)
plt.show()
print(img)
#Save pictures
imageio.imsave('test.jpg',img)

 

To be continued