<p>OpenCV的FFT函数要求将缩放标志<code>DFT_SCALE</code>恰好指定给<code>dft,idft</code>中的一个。见<a href="https://docs.opencv.org/3.4/d2/de8/group__core__array.html#gaa708aa2d2e57a508f968eb0f69aa5ff1" rel="nofollow noreferrer">https://docs.opencv.org/3.4/d2/de8/group__core__array.html#gaa708aa2d2e57a508f968eb0f69aa5ff1</a>处的说明</p>
<p>我选择在第一个函数中传递它,因为它已经接受了另一个标志</p>
<p>numpy的fft2函数不需要这个。他们按原样工作</p>
<p>这里是两个版本,只是在<code>dft/idft</code>和<code>fft2/ifft2</code>上有所不同</p>
<pre class="lang-py prettyprint-override"><code>import os
import sys
import numpy as np
import cv2 as cv
def prepare_spectrum(spectrum):
mag = np.linalg.norm(spectrum, axis=2)
mag /= mag.max()
mag **= 1/4
return mag
img = cv.imread(cv.samples.findFile("lena.jpg"), cv.IMREAD_GRAYSCALE)
img = img / np.float32(255)
dft = cv.dft(img, flags=cv.DFT_COMPLEX_OUTPUT | cv.DFT_SCALE)
cv.imshow("1 dft", prepare_spectrum(dft))
dft_shift = np.fft.fftshift(dft)
cv.imshow("2 dft_shift", prepare_spectrum(dft_shift))
rows, cols = img.shape
crow,ccol = (rows // 2), (cols // 2)
mask = np.zeros((rows,cols,2), dtype=np.uint8)
mask[crow-30:crow+30, ccol-30:ccol+30] = 1
dft_shift *= mask
cv.imshow("3 dft_shift masked", prepare_spectrum(dft_shift))
dft = np.fft.ifftshift(dft_shift)
cv.imshow("4 dft masked", prepare_spectrum(dft))
img_back = cv.idft(dft)
img_back = img_back[:,:,0] # real part
cv.imshow('before', img)
cv.imshow('after', img_back)
cv.waitKey()
cv.destroyAllWindows()
</code></pre>
<p>努比:</p>
<pre class="lang-py prettyprint-override"><code>import os
import sys
import numpy as np
import cv2 as cv
def prepare_spectrum(spectrum):
mag = np.abs(spectrum)
mag /= mag.max()
mag **= 1/4
return mag
img = cv.imread(cv.samples.findFile("lena.jpg"), cv.IMREAD_GRAYSCALE)
img = img / np.float32(255)
dft = np.fft.fft2(img)
cv.imshow("1 dft", prepare_spectrum(dft))
dft_shift = np.fft.fftshift(dft)
cv.imshow("2 dft_shift", prepare_spectrum(dft_shift))
rows, cols = img.shape
crow,ccol = (rows // 2), (cols // 2)
mask = np.zeros((rows,cols), dtype=np.uint8)
mask[crow-30:crow+30, ccol-30:ccol+30] = 1
dft_shift *= mask
cv.imshow("3 dft_shift masked", prepare_spectrum(dft_shift))
dft = np.fft.ifftshift(dft_shift)
cv.imshow("4 dft masked", prepare_spectrum(dft))
img_back = np.fft.ifft2(dft)
img_back = np.real(img_back)
cv.imshow('before', img)
cv.imshow('after', img_back)
cv.waitKey()
cv.destroyAllWindows()
</code></pre>
<p><a href="https://i.stack.imgur.com/f5CS7m.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/f5CS7m.png" alt="spectrum masked"/></a>
<a href="https://i.stack.imgur.com/hZOERm.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/hZOERm.png" alt="lowpassed lena"/></a></p>