def _process(self, psf):
w = self.display.widget
img = w.image
img = img[w.currentIndex].copy()
#Wiener method needs image scaled (-1,1):
mn, mx = img.min(), img.max()
img-=mn
img/=(mx-mn)
vals = self._bRange()
out = []
for balance in vals:
o = wiener(img, psf, balance)
#Scale back:
o[o<0]=0
o*=(mx-mn)
o+=mn
out.append(o)
return out, ['Balance=%s' %v for v in vals]
def _process(self, psf):
w = self.display.widget
img = w.image
img = img[w.currentIndex].copy()
#Wiener method needs image scaled (-1,1):
mn, mx = img.min(), img.max()
img -= mn
img /= (mx - mn)
vals = self._bRange()
out = []
for balance in vals:
o = wiener(img, psf, balance)
#Scale back:
o[o < 0] = 0
o *= (mx - mn)
o += mn
out.append(o)
return out, ['Balance=%s' % v for v in vals]
def _correctBlur(self, image, light_spectrum, date):
#save startup time
from skimage.restoration.deconvolution import unsupervised_wiener, wiener
d = self.getCoeff('psf', light_spectrum, date)
if not d:
print 'skip deconvolution // no PSF set'
return image
psf = d[2]
mx = image.max()
image/=mx
balance = self.getCoeff('balance', light_spectrum, date)
if balance is None:
print 'no balance value for wiener deconvolution found // use unsupervised_wiener instead // this will take some time'
deconvolved, _ = unsupervised_wiener(image, psf)
else:
deconvolved = wiener(image, psf, balance[2])
deconvolved[deconvolved<0]=0
deconvolved*=mx
return deconvolved
def _correctBlur(self, image, light_spectrum, date):
#save startup time
from skimage.restoration.deconvolution import unsupervised_wiener, wiener
d = self.getCoeff('psf', light_spectrum, date)
if not d:
print 'skip deconvolution // no PSF set'
return image
psf = d[2]
mx = image.max()
image /= mx
balance = self.getCoeff('balance', light_spectrum, date)
if balance is None:
print 'no balance value for wiener deconvolution found // use unsupervised_wiener instead // this will take some time'
deconvolved, _ = unsupervised_wiener(image, psf)
else:
deconvolved = wiener(image, psf, balance[2])
deconvolved[deconvolved < 0] = 0
deconvolved *= mx
return deconvolved
def inv_filter(img, S, noise=0.001):
# If you want to call inv_filter, you have to impute first to avoid
# spectral artifacts
assert(np.all(img.mask == False))
# unfortunately scipy deconvolution messes with our scale...
F = pad2(img)
# Do my scaling here to keep scipy happy!
machine_scale = F.max()
F /= machine_scale
S = S / S.sum() # avoids any scaling effects
img_w, img_h, img_ch = img.shape
out = np.zeros_like(F)
for ch in range(img_ch):
out[:, :, ch] = deconvolution.wiener(F[:, :, ch].data.astype(float),
S, noise)
out = machine_scale * out[:img_w, :img_h]
# out[img.mask] = 0.
return np.ma.masked_array(out, mask=img.mask)