灰度指纹图像的归一化

Question

我正在尝试按照第 2.3 节中的this paper 中的建议对灰度指纹图像进行标准化。为了测试它，我使用了这个gray fingerprint，标准化后我希望它看起来像这样normalized fingerprint。

这是我用来标准化指纹的函数

import numpy as np
from glob import glob
from PIL import Image
import matplotlib.pyplot as plt


def normalization(im_arr, M0, VAR0):
    """
    im_arr: image as array
    M0: desired mean
    VAR0: desired variance

    G: normalized image as array
    """
    row, col = im_arr.shape
    # Mean and variance calculation
    M = (1 / (row ** 2)) * np.sum(im_arr)
    VAR = (1 / (row ** 2)) * np.sum((im_arr - M) ** 2)

    G = np.zeros(im_arr.shape)
    for i in range(row):
        for j in range(col):
            if im_arr[i, j] > M:
                G[i, j] = M0 + np.sqrt((VAR0 * (im_arr[i, j] - M) ** 2) / VAR)
            else:
                G[i, j] = M0 - np.sqrt((VAR0 * (im_arr[i, j] - M) ** 2) / VAR)
    return G

这是我用来读取数据的代码

data = 'greyfp.JPG'

img = Image.open(data).convert('L')
img = img.resize((200, 200))
img_arr = np.array(img).astype(float)
img_arr = img_arr / np.max(img_arr)

G = normalization(img_arr, 0.5, 0.5)

如果我现在检查新图像的均值和方差，我会得到想要的结果。但是当我尝试从其矩阵中绘制归一化指纹时，它看起来与输入图像完全相同

fig, ax = plt.subplots(1, 2)
grey = ax[0].imshow(img, cmap='gray')
norm = ax[1].imshow(G, cmap='gray')
plt.show()

但是，如果我保存新图像并绘制它

img = Image.fromarray(G, 'L')
img.save('G.JPG')
plt.imshow(img, cmap='gray')
plt.show()

它看起来 completely different，但仍然不像我想要的输出。

谁能帮我得到与论文中相同的结果，也许可以帮助我理解，为什么矩阵和图像类型的图看起来完全不同？

Answer 1

查看您的代码更新：

import numpy as np
from glob import glob
from PIL import Image
import matplotlib.pyplot as plt


def normalization(im_arr, M0, VAR0):
    """
    im_arr: image as array
    M0: desired mean
    VAR0: desired variance

    G: normalized image as array
    """
    row, col = im_arr.shape
    print("row,col = ", row, col)
    # Mean and variance calculation
    M = (1 / (row ** 2)) * np.sum(im_arr)
    VAR = (1 / (row ** 2)) * np.sum((im_arr - M) ** 2)

    G = np.zeros(im_arr.shape)
    for i in range(row):
        for j in range(col):

            if im_arr[i, j] > M:
                G[i, j] = M0 + np.sqrt((VAR0 * (im_arr[i, j] - M) ** 2) / VAR)
            else:
                G[i, j] = M0 - np.sqrt((VAR0 * (im_arr[i, j] - M) ** 2) / VAR)
    return G

data = 'test.jpg'

img = Image.open(data).convert('L')
img = img.resize((200, 200))
img_arr = np.array(img).astype(float)
img_arr = img_arr / np.max(img_arr)

G = normalization(img_arr, 0.5, 0.5)
G=G*255
G=np.where(G<=255,G,255)
G=np.array(G).astype(np.uint8)
cv2.imshow("G",G)


img = Image.fromarray(G, 'L')
img.save('G.JPG')
plt.imshow(img, cmap='gray')
plt.show()