网上看到一篇文章，关于深度学习预处理，提取图片的高频通道进行训练，有利于结果准确率的提升和抗干扰
于是对图片的高频通道进行了解和学习
1. 傅里叶变换实现高通滤波（单通道图像）
# -*- coding: utf-8 -*-
import cv2 as cv
import numpy as np
from matplotlib import pyplot as plt
#读取图像：此处图像为单通道图像
img = cv.imread('Lena.png', 0)
#傅里叶变换
f = np.fft.fft2(img)
fshift = np.fft.fftshift(f)
#设置高通滤波器
rows, cols = img.shape
crow,ccol = int(rows/2), int(cols/2)
fshift[crow-30:crow+30, ccol-30:ccol+30] = 0
#傅里叶逆变换
ishift = np.fft.ifftshift(fshift)
iimg = np.fft.ifft2(ishift)
iimg = np.abs(iimg)
#显示原始图像和高通滤波处理图像
plt.subplot(121), plt.imshow(img, 'gray'), plt.title('Original Image')
plt.axis('off')
plt.subplot(122), plt.imshow(iimg, 'gray'), plt.title('Result Image')
plt.axis('off')
plt.show()
傅里叶变换只能用于单通道图像
对于常规的三通道图像，需要进行变换和处理
2. 图像通道的分离与合并
import numpy as np;
import cv2; #导入opencv模块
image=cv2.imread("/home/zje/Pictures/lena.jpeg");#读取要处理的图片
B,G,R = cv2.split(image); #分离出图片的B，R，G颜色通道
zeros = np.zeros(image.shape[:2],dtype="uint8");#创建与image相同大小的零矩阵
cv2.imshow("BLUE",cv2.merge([B,zeros,zeros]));#显示 （B，0，0）图像
cv2.imshow("GREEN",cv2.merge([zeros,G,zeros]));#显示（0，G，0）图像
cv2.imshow("RED",cv2.merge([zeros,zeros,R]));#显示（0，0，R）图像
cv2.waitKey(0);
3. 傅里叶变换实现高通滤波
# -*- coding: utf-8 -*-
import cv2
import numpy as np
from matplotlib import pyplot as plt
def Fourier_high_pass(img, offset):
#傅里叶变换
# dft = cv2.dft(np.float32(img), flags = cv2.DFT_COMPLEX_OUTPUT)
# fshift = np.fft.fftshift(dft)
f = np.fft.fft2(img)
fshift = np.fft.fftshift(f)
#设置高通滤波器
rows, cols = img.shape
crow, ccol = int(rows / 2), int(cols / 2)
fshift[crow - offset:crow + offset, ccol - offset:ccol + offset] = 0
#傅里叶逆变换
ishift = np.fft.ifftshift(fshift)
iimg = np.fft.ifft2(ishift)
iimg = np.abs(iimg)
# 频谱图像双通道复数转换为0-255区间
# res = cv2.magnitude(iimg[:, :, 0], iimg[:, :, 1])
res = iimg
res = 255 * (res - np.min(res)) / (np.max(res) - np.min(res))
return res
image_path = "2_36.0.jpg"
image = cv2.imread(image_path)
B, G, R = cv2.split(image)
zeros = np.zeros(image.shape[:2], dtype="float32")
for offset in range(10, 101, 10):
res_b = Fourier_high_pass(B, offset)
# cv2.imwrite("res_b.jpg", cv2.merge([res_b, zeros, zeros]))
res_g = Fourier_high_pass(G, offset)
# cv2.imwrite("res_g.jpg", cv2.merge([zeros, res_g, zeros]))
res_r = Fourier_high_pass(R, offset)
# cv2.imwrite("res_r.jpg", cv2.merge([zeros, zeros, res_r]))
res_merge = cv2.merge([res_b, res_g, res_r])
cv2.imwrite("res_high_"+str(offset)+".jpg", res_merge)
4. 傅里叶变换实现低通滤波
# -*- coding: utf-8 -*-
import cv2
import numpy as np
from matplotlib import pyplot as plt
def Fourier_low_pass(img, offset):
#傅里叶变换
dft = cv2.dft(np.float32(img), flags = cv2.DFT_COMPLEX_OUTPUT)
fshift = np.fft.fftshift(dft)
#设置低通滤波器
rows, cols = img.shape
crow,ccol = int(rows/2), int(cols/2) #中心位置
mask = np.zeros((rows, cols, 2), np.uint8)
mask[crow-offset:crow+offset, ccol-offset:ccol+offset] = 1
#掩膜图像和频谱图像乘积
f = fshift * mask
# print(f.shape, fshift.shape, mask.shape)
#傅里叶逆变换
ishift = np.fft.ifftshift(f)
iimg = cv2.idft(ishift)
# 频谱图像双通道复数转换为0-255区间
res = cv2.magnitude(iimg[:, :, 0], iimg[:, :, 1])
res = 255 * (res - np.min(res)) / (np.max(res) - np.min(res))
return res
image_path = "2_36.0.jpg"
image = cv2.imread(image_path)
B, G, R = cv2.split(image)
zeros = np.zeros(image.shape[:2], dtype="float32")
for offset in range(10, 101, 10):
res_b = Fourier_low_pass(B, offset)
# cv2.imwrite("res_b.jpg", cv2.merge([res_b, zeros, zeros]))
res_g = Fourier_low_pass(G, offset)
# cv2.imwrite("res_g.jpg", cv2.merge([zeros, res_g, zeros]))
res_r = Fourier_low_pass(R, offset)
# cv2.imwrite("res_r.jpg", cv2.merge([zeros, zeros, res_r]))
res_merge = cv2.merge([res_b, res_g, res_r])
cv2.imwrite("res_low_"+str(offset)+".jpg", res_merge)
参考资料：
0. CMU团队解析CNN泛化能力：一切秘密都在数据中
1. [Python图像处理] 二十二.Python图像傅里叶变换原理及实现
2. [Python图像处理] 二十三.傅里叶变换之高通滤波和低通滤波
3. python3+opencv 图像通道的分离（split（）函数）和合并（merge（）函数）
4. Python将二维数组归一化到0-255
