python - 在 Python 中手动高效地创建图像直方图

Question

我想编写无需使用内置 Matplotlib hist 函数即可显示图像直方图的代码。

这是我的代码:

import cv2 as cv
import numpy as np
from matplotlib import pyplot as plt

def manHist(img):
   row, col = img.shape # img is a grayscale image
   y = np.zeros((256), np.uint64)
   for i in range(0,row):
      for j in range(0,col):
         y[img[i,j]] += 1
   x = np.arange(0,256)
   plt.bar(x,y,color="gray",align="center")
   plt.show()

def main():
   img = cv.imread("C:/Users/Kadek/Documents/MATLAB/2GS.jpg")
   manHist(img)

main()

我的问题是，有没有更有效的方法来制作像素值频率数组而不使用 for 循环？

Answer 1

基于 NumPy 的矢量化解决方案将使用 np.bincount -

out = np.bincount(img.ravel(),minlength=256)

另一种基于 .sum() 的向量化方法-

out = (img.ravel() == np.arange(256)[:,None]).sum(1)

sample 运行以验证结果 -

In [155]: # Input image (512x512) as array
     ...: img = np.random.randint(0,255,(512,512))
     ...: 
     ...: # Original code
     ...: row, col = img.shape
     ...: y = np.zeros((256), np.uint64)
     ...: for i in range(0,row):
     ...:     for j in range(0,col):
     ...:         y[img[i,j]] += 1
     ...:         

In [156]: out1 = np.bincount(img.ravel(),minlength=256)

In [157]: out2 = (img.ravel() == np.arange(256)[:,None]).sum(1)

In [158]: np.allclose(y,out1)
Out[158]: True

In [159]: np.allclose(y,out2)
Out[159]: True