python中无符号整型数据的减法运算

一、问题描述

---

cv2.imread读入图像，并显示读入图像间的残差热图，获取的结果明显不是我们想要的结果。

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

dirpath = r"C:/Users/zhang/Desktop/testResult/ImageQuality/Image/DuiBi49_13/"
CBCT0 = cv2.imread(dirpath + 'CBCT.png')
PCT0 = cv2.imread(dirpath + 'PCT.png')
SPCT0 = cv2.imread(dirpath + 'MindCg.png')

CPerr = CBCT-PCT
SPerr = SPCT-PCT

热图中显示的像素值有大量的200以上的像素值,与我预期想得到的很不一样。

二、解决

---

查看读取的图像的数据类型是uint8类型的，无符号整型的数据求差，我们将问题简化一下：

import numpy as np

a = np.uint8([1,2,3,4,5])
b = np.uint8([5,4,3,2,1])
e = a - b
print(e)

 这[252,254,0,2,4]显然不是我们想要的结果[-4,-2,02,4]，主要是因为无符号整型的运算：就是那些源码、反码、补码的那些知识。那我们可以将a,b转换为有符号整型吗:

import numpy as np

a = np.uint8([1,2,3,4,5])
b = np.uint8([5,4,3,2,1])
c = int(a)
b = int(b)
e = c - d
print(e)

 没有查询到python中的无符号整型如何强制类型转换成有符号数，但是我们分情况处理仍然能解决，记录一下：

import numpy as np

a = np.uint8([1,2,3,4,5])
b = np.uint8([5,4,3,2,1])

n = len(a)
e = [1]*n
for i in range(n):
    if a[i] >= b[i]:
        e[i] = a[i] - b[i]
    else:
        e[i] = (a[i] -b[i]) - 256
print(e)

 应用到开始的图像上：

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

dirpath = r"C:/Users/zhang/Desktop/testResult/ImageQuality/Image/DuiBi49_13/"
CBCT0 = cv2.imread(dirpath + 'CBCT.png')
PCT0 = cv2.imread(dirpath + 'PCT.png')
SPCT0 = cv2.imread(dirpath + 'MindCg.png')

CPerr = [[1 for i in range(256)] for i in range(256)]
SPerr = [[1 for i in range(256)] for i in range(256)]
for i in range(256):
    for j in range(256):
        if CBCT[i][j] >= PCT[i][j]:
            CPerr[i][j] = CBCT[i][j] - PCT[i][j]
        else:
            CPerr[i][j] = (CBCT[i][j] - PCT[i][j]) - 256

for i in range(256):
    for j in range(256):
        if SPCT[i][j] >= PCT[i][j]:
            SPerr[i][j] = SPCT[i][j] - PCT[i][j]
        else:
            SPerr[i][j] = (SPCT[i][j] - PCT[i][j]) - 256

cmap = sns.diverging_palette(0,230,90,60,as_cmap=True)
plt.subplot(2,3,1)
plt.imshow(PCT0)
plt.subplot(2,3,2)
plt.imshow(CBCT0)
plt.subplot(2,3,3)
plt.imshow(SPCT0)
plt.subplot(2,2,3)
sns.heatmap(CPerr,vmin=-100,vmax=100,cmap=cmap)
plt.subplot(2,2,4)
sns.heatmap(SPerr,vmin=-100,vmax=100,cmap=cmap)
plt.tight_layout()
plt.show()