def phase_correlation(a, b):
G_a = np.fft.fft2(a)
G_b = np.fft.fft2(b)
conj_b = np.ma.conjugate(G_b)
R = G_a*conj_b
R /= np.absolute(R)
r = np.fft.ifft2(R).real
return r
Here is an example: We take two similar images, but of different phases and plot the phase correlation (a black image with a single white dot at the appropriate phase difference).
from scipy import misc
from matplotlib import pyplot
import numpy as np
#Get two images with snippet at different locations
im1 = np.mean(misc.face(), axis=-1) #naive colour flattening
im2 = np.zeros_like(im1)
im2[:200,:200] = im1[200:400, 500:700]
corrimg = phase_correlation(im1, im2)
r,c = np.unravel_index(corrimg.argmax(), corrimg.shape)
pyplot.imshow(im1)
pyplot.plot([c],[r],'ro')
pyplot.show()
pyplot.imshow(im2)
pyplot.show()
pyplot.figure(figsize=[8,8])
pyplot.imshow(corrimg, cmap='gray')
pyplot.show()
from pylab import *
corr = (ifftn(fftn(a)*ifftn(b))).real
imshow(corr)
using scipy this should be a one-liner (although you can probably avoid the ndimage package)
from scipy.fftpack import fftn, ifftn
corr = (ifftn(fftn(a)*ifftn(b))).real
assuming you've managed to read your original images into numpy arrays a & b. If it's 2D images mayavi might be a bit overkill, and it would probably be easier to use matplotlib than chaco. If using matplotlib, you could do the whole lot with
from pylab import *
corr = (ifftn(fftn(a)*ifftn(b))).real
imshow(corr)
发布评论
评论（3）
相位相关性如http://en.wikipedia.org/wiki/Phase_correlation所述，摘自https://github.com/michaelting/Phase_Correlation/blob/master/phase_corr.py 。
这是一个示例：我们拍摄两个相似的图像，但相位不同，并绘制相位相关性（黑色图像，在适当的相位差处有一个白点）。
Phase correlation as described by http://en.wikipedia.org/wiki/Phase_correlation, taken from https://github.com/michaelting/Phase_Correlation/blob/master/phase_corr.py.
Here is an example: We take two similar images, but of different phases and plot the phase correlation (a black image with a single white dot at the appropriate phase difference).
使用 scipy 这应该是一行（尽管你可能可以避免使用 ndimage 包）
假设你已经成功地将原始图像读入 numpy 数组 a & 中。 b.如果是 2D 图像，mayavi 可能有点大材小用，而且使用 matplotlib 可能比 chaco 更容易。如果使用 matplotlib，你可以用
using scipy this should be a one-liner (although you can probably avoid the ndimage package)
assuming you've managed to read your original images into numpy arrays a & b. If it's 2D images mayavi might be a bit overkill, and it would probably be easier to use matplotlib than chaco. If using matplotlib, you could do the whole lot with
Scipy 在其 scipy.ndimage 包中包含许多图像处理例程。
Scipy contains many image processing routines in its scipy.ndimage package.