Question

我用过公式：（（L-1）/ MN）的 NI 哪里 L是graylevels的总数，M N是图像的大小，ni是累积频率

但我总是得到完整的黑色图像。我也尝试过其他图像。

import numpy as np
import cv2

path="C:/Users/Arun Nambiar/Downloads/fingerprint256by256 (1).pgm"
img=cv2.imread(path,0)

#To display image before equalization
cv2.imshow('image',img)
cv2.waitKey(0)
cv2.destroyAllWindows()


a=np.zeros((256,),dtype=np.float16)
b=np.zeros((256,),dtype=np.float16)
height,width=img.shape

#finding histogram
for i in range(width):
    for j in range(height):
    g=img[j,i]
    a[g]=a[g]+1
print(a)        


#performing histogram equalization

tmp=255/(height*width)

a[0]=tmp*a[0]
b[0]=round(a[0])

for g in range(1,width):
   a[g]=(a[g]*tmp)+(a[g-1]*tmp)
   b[g]=round(a[g])



print(b)


b=b.astype(np.uint8)
print(b)
for i in range(width):
    for j in range(height):
        g=img[j,i]
        img[j,i]=b[g]

cv2.imshow('image',img)
cv2.waitKey(0)
cv2.destroyAllWindows()

我附加了图片this is image i have used

Answer 1

均衡步骤的实施有些不正确。概率分布函数（PDF）的计算应该高达箱的数量而不是图像宽度（尽管在这种特定情况下它们是相等的）。请参阅以下代码，并更正均衡步骤。

import numpy as np
import cv2

path = "fingerprint256by256.pgm"
img = cv2.imread(path,0)

#To display image before equalization
cv2.imshow('image',img)
cv2.waitKey(0)


a = np.zeros((256,),dtype=np.float16)
b = np.zeros((256,),dtype=np.float16)

height,width=img.shape

#finding histogram
for i in range(width):
    for j in range(height):
        g = img[j,i]
        a[g] = a[g]+1

print(a)   


#performing histogram equalization
tmp = 1.0/(height*width)
b = np.zeros((256,),dtype=np.float16)

for i in range(256):
    for j in range(i+1):
        b[i] += a[j] * tmp;
    b[i] = round(b[i] * 255);

# b now contains the equalized histogram
b=b.astype(np.uint8)

print(b)

#Re-map values from equalized histogram into the image
for i in range(width):
    for j in range(height):
        g = img[j,i]
        img[j,i]= b[g]

cv2.imshow('image',img)
cv2.waitKey(0)
cv2.destroyAllWindows()

使用Python 3.4和OpenCV 3.4在Ubuntu 14.04上测试和验证。

使用python和opencv进行直方图均衡，而不使用内置函数

1 个答案: