def test_shifts(xsh,ysh,imsize,gaussfit):
image,new_image,tolerance = make_offset_images(xsh, ysh, imsize)
if gaussfit:
xoff,yoff,exoff,eyoff = cross_correlation_shifts(image,new_image,return_error=True)
print xoff,yoff,np.abs(xoff-xsh),np.abs(yoff-ysh),exoff,eyoff
else:
xoff,yoff = chi2_shift(image,new_image,return_error=False)
print xoff,yoff,np.abs(xoff-xsh),np.abs(yoff-ysh)
assert np.abs(xoff-xsh) < tolerance
assert np.abs(yoff-ysh) < tolerance
def compare_methods(im1,im2, ntests=100, noise=np.std,
usfac=201, **kwargs):
"""
Perform tests with noise added to determine the errors on the
'best-fit' offset
Parameters
----------
usfac : int
upsampling factor; governs accuracy of fit (1/usfac is best accuracy)
ntests : int
Number of tests to run
noise : func or real
Either a function to apply to im2 to determine the noise to use, or
a fixed noise level
"""
try:
noise = noise(im2)
except TypeError:
pass
try:
import progressbar
widgets = [progressbar.FormatLabel('Processed: %(value)d offsets in %(elapsed)s'), progressbar.Percentage()]
progress = progressbar.ProgressBar(widgets=widgets)
except ImportError:
def progress(x):
yield x
offsets = []
eoffsets = []
for test_number in progress(xrange(ntests)):
extra_noise = np.random.randn(*im2.shape) * noise
dxr, dyr, edxr, edyr = register_images(im1, im2+extra_noise, usfac=usfac,
return_error=True, **kwargs)
dxccs, dyccs, edxccs, edyccs = cross_correlation_shifts(im1,
im2+extra_noise,
errim2=im2*0+extra_noise.std(),
return_error=True, **kwargs)
# too slow!!!
dxccg, dyccg, edxccg, edyccg = 0,0,0,0
#dxccg, dyccg, edxccg, edyccg = cross_correlation_shifts(im1,
# im2+extra_noise, return_error=True, gaussfit=True,
# **kwargs)
dxchi, dychi, edxchi, edychi = chi2_shift(im1, im2+extra_noise,
err=im2*0+noise,
return_error=True, upsample_factor='auto', verbose=False, **kwargs)
offsets.append([dxr,dyr,dxccs,dyccs,dxccg,dyccg,dxchi,dychi])
eoffsets.append([edxr,edyr,edxccs,edyccs,edxccg,edyccg,edxchi,edychi])
return np.array(offsets),np.array(eoffsets)
def run_tests(xsh, ysh, imsize, amp, gaussfit, nfits=1000, maxoff=20):
fitted_shifts = np.array(do_n_fits(nfits, xsh, ysh, imsize, amp=amp, maxoff=maxoff))
errors = fitted_shifts.std(axis=0)
x,y,ex,ey = cross_correlation_shifts(
*make_offset_images(xsh, ysh, imsize, amp=amp)[:2],
gaussfit=gaussfit, maxoff=maxoff, return_error=True,
errim1=np.ones([imsize,imsize]),
errim2=np.ones([imsize,imsize]))
print "StdDev: %10.3g,%10.3g Measured: %10.3g,%10.3g "+\
" Difference: %10.3g, %10.3g Diff/Real: %10.3g,%10.3g" % (
errors[0],errors[1], ex,ey,errors[0]-ex,errors[1]-ey,
(errors[0]-ex)/errors[0], (errors[1]-ey)/errors[1])
return errors[0],errors[1],ex,ey
