wassname/scikit-image
1
0
Fork 0
mirror of https://github.com/wassname/scikit-image.git synced 2026-07-05 10:00:13 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix the bad results of unsupervised deconvolution.

Browse Source 
The initial implementation was simplified to reduce the number of
local variable. Consequently the parameter data.size was wrong. It was
the size of the fourier spectrum, with hermitian property, instead of
the real space size. This should work.

In addition the PSF is now normalized.
This commit is contained in:
François Orieux
2013-10-28 14:50:05 +01:00
parent 4b233098e7
commit 7042d778e6
4 changed files with 7 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
skimage/deconvolution/tests/camera_unsup.npy
BIN
View File
Binary file not shown.
skimage/deconvolution/tests/camera_wiener.npy
BIN
View File
Binary file not shown.
skimage/deconvolution/tests/test_deconvolution.py
+2 -2
View File
@@ -8,7 +8,7 @@ test_img = camera().astype(np.float)
def test_wiener():
psf = np.ones((5, 5))
psf = np.ones((5, 5)) / 25
data = convolve2d(test_img, psf, 'same')
np.random.seed(0)
data += 0.1 * data.std() * np.random.standard_normal(data.shape)
@@ -19,7 +19,7 @@ def test_wiener():
def test_unsupervised_wiener():
psf = np.ones((5, 5))
psf = np.ones((5, 5)) / 25
data = convolve2d(test_img, psf, 'same')
np.random.seed(0)
data += 0.1 * data.std() * np.random.standard_normal(data.shape)
skimage/deconvolution/wiener.py
+5 -4
View File
@@ -82,7 +82,7 @@ def wiener(data, psf, reg_val, reg=None, real=True):
>>> from skimage import color, data, deconvolution
>>> lena = color.rgb2gray(data.lena())
>>> from scipy.signal import convolve2d
>>> psf = np.ones((5, 5))
>>> psf = np.ones((5, 5)) / 25
>>> lena = convolve2d(lena, psf, 'same')
>>> lena += 0.1 * lena.std() * np.random.standard_normal(lena.shape)
>>> deconvolved_lena = deconvolution.wiener(lena, psf, 1100)
@@ -212,7 +212,7 @@ def unsupervised_wiener(data, psf, reg=None, user_params=None):
>>> from skimage import color, data, deconvolution
>>> lena = color.rgb2gray(data.lena())
>>> from scipy.signal import convolve2d
>>> psf = np.ones((5, 5))
>>> psf = np.ones((5, 5)) / 25
>>> lena = convolve2d(lena, psf, 'same')
>>> lena += 0.1 * lena.std() * np.random.standard_normal(lena.shape)
>>> deconvolved_lena = deconvolution.unsupervised_wiener(lena, psf)
@@ -258,6 +258,7 @@ def unsupervised_wiener(data, psf, reg=None, user_params=None):
areg2 = np.abs(reg)**2
atf2 = np.abs(trans_fct)**2
data_size = data.size
data = uft.urfft2(data.astype(np.float))
# Gibbs sampling
@@ -279,12 +280,12 @@ def unsupervised_wiener(data, psf, reg=None, user_params=None):
params['callback'](x_sample)
# sample of Eq. 31 p(gn | x^k, gx^k, y)
gn_chain.append(npr.gamma(data.size / 2,
gn_chain.append(npr.gamma(data_size / 2,
2 / uft.image_quad_norm(data - x_sample *
trans_fct)))
# sample of Eq. 31 p(gx | x^k, gn^k-1, y)
gx_chain.append(npr.gamma((data.size - 1) / 2,
gx_chain.append(npr.gamma((data_size - 1) / 2,
2 / uft.image_quad_norm(x_sample * reg)))
# current empirical average