[opencv learning] [image histogram and equalization]

Today, learn image histogram and image equalization

One ： Image histogram

import cv2
import numpy as np
import matplotlib.pyplot as plt

#  Histogram of the image , It is to count which pixels , Or the number of pixels in that pixel range 
#  stay opencv Is used in  calcHist(images, channels, mask, histSzie, ranges)
#  Argument parsing ：
# images:  Multiple images can be transferred in at one time . The incoming image is a list,  use [] Cover up , The data type is  uint8/float32
# channels:  Which channel is counted , If the grayscale image is imported [0], If it's a color image , Pass in [0]、[1]、[2], It stands for  BGR passageway 
# mask:  Mask the image , If you want to count this image , Is the incoming None,  If you want to count only a part , You need to make a mask to pass in 
# histSzie: BIN The number of , Divide into several ranges on average , If it is a range of pixels , Namely 256.
# ranges:  Range of pixels , Usually it is [0, 256]

#  Show the image , Encapsulate as a function 
def cv_show_image(name, img):
    cv2.imshow(name, img)
    cv2.waitKey(0)  #  Waiting time , In milliseconds ,0 Represents any key termination 
    cv2.destroyAllWindows()

img = cv2.imread('images/saoge2.jpg')
print(img.shape)
plt.hist(img.ravel(), 256)
plt.title('src_img_hist')
plt.show()

#  Calculate the histogram of each channel 
hist_b = cv2.calcHist([img], [0], None, [256], [0, 256])
print(hist_b.shape)  # (256, 1)
hist_g = cv2.calcHist([img], [1], None, [256], [0, 256])
print(hist_g.shape)  # (256, 1)
hist_r = cv2.calcHist([img], [2], None, [256], [0, 256])
print(hist_r.shape)  # (256, 1)

#  Draw this BGR Histogram of three channels 
plt.plot(hist_b, 'b')
plt.plot(hist_g, 'g')
plt.plot(hist_r, 'r')
plt.xlim([0, 256])
plt.title('bgr_hist')
plt.show()

#  With mask operation , When we only want to take part of the region for histogram statistics , You can use the mask 
#  Mask is the selected area is reserved , The unselected area is abandoned , Therefore, the value of the mask in the selection area is 255, The value of the discarded area is 0
#  Here's the definition mask The way of mask 
mask = np.zeros(img.shape[:2], np.uint8)
mask[100:300, 100:540] = 255

masked_img = cv2.bitwise_and(img, img, mask=mask)  #  Get the image after flax action 

hist_full_img = cv2.calcHist([img], [0], None, [256], [0, 256])
hist_mask_img = cv2.calcHist([img], [0], mask, [256], [0, 256])

plt.subplot(2,2,1), plt.imshow(img, 'gray')
plt.subplot(2,2,2), plt.imshow(mask, 'gray')
plt.subplot(2,2,3), plt.imshow(masked_img, 'gray')
plt.subplot(2,2,4), plt.plot(hist_full_img, 'b'), plt.plot(hist_mask_img, 'g')
plt.xlim([0, 256])
plt.title('mask_hist')
plt.show()

The gray histogram of the original image is ：

Of the original image BGR Histogram of three channels

Of the original image B Masked B Channel histogram

Two ： Histogram equalization

#  Histogram equalization 
#  After getting the histogram , Some histograms have uneven pixel distribution , The color is very biased , Therefore, it needs to be balanced and adjusted 
#  The purpose of image equalization is to improve the contrast of the image . For the image with concentrated gray distribution （ The whole is too dark or too bright ）, Can play a balanced role （ There is light and dark ）.
#  The way is ：
#  Read image 
#  Count the pixel values for each channel separately [0,255] Number of occurrences .
#  Calculate the pixel value of each channel separately [0,255] Probability of occurrence , Get the probability histogram .
#  Calculate the pixel value of each channel separately [0,255] Cumulative sum of probabilities , Get the cumulative probability histogram .
#  Calculate the pixel mapping function for each channel according to the cumulative probability histogram . Complete the mapping of each pixel for each channel .
#  for instance , The pixel value is 56 The cumulative probability of is 0.2, Then the pixel value is 56 The new mapped pixel value of is  256*0.2=51
#  for instance , The pixel value is 255 The cumulative probability of must be 1, The pixel value is 255 The new mapped pixel value of is  256*01=255
#  Output histogram equalization image .

#  stay opencv The operation 
#  Read the original image 
img = cv2.imread('images/saoge2.jpg', cv2.IMREAD_GRAYSCALE)  #  Read gray image 
plt.hist(img.ravel(), 256)
plt.title('src_img_hist')
plt.show()

#  Image histogram equalization , The whole image will be equalized 
equ = cv2.equalizeHist(img)
plt.hist(equ.ravel(), 256)
plt.title('equalizeHist_img')
plt.show()

#  Adaptive equalization 
#  Divide the image into small squares , Do their own equalization 
clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))
res_clahe = clahe.apply(img)

res = np.hstack((img, equ, res_clahe))
cv_show_image('equalizeHist', res)

Comparison of the effects of general equalization and adaptive equalization