def create_box_convolutionfunction(im, oversampling=1, support=1):
""" Fill a box car function into a ConvolutionFunction
Also returns the griddata correction function as an image
:param im: Image template
:param oversampling: Oversampling of the convolution function in uv space
:return: griddata correction Image, griddata kernel as ConvolutionFunction
"""
assert isinstance(im, Image)
cf = create_convolutionfunction_from_image(im, oversampling=1, support=4)
nchan, npol, _, _ = im.shape
cf.data[...] = 0.0 + 0.0j
cf.data[..., 2, 2] = 1.0 + 0.0j
# Now calculate the griddata correction function as an image with the same coordinates as the image
# which is necessary so that the correction function can be applied directly to the image
nchan, npol, ny, nx = im.data.shape
nu = numpy.abs(coordinates(nx))
gcf1d = numpy.sinc(nu)
gcf = numpy.outer(gcf1d, gcf1d)
gcf = 1.0 / gcf
gcf_data = numpy.zeros_like(im.data)
gcf_data[...] = gcf[numpy.newaxis, numpy.newaxis, ...]
gcf_image = create_image_from_array(gcf_data, cf.projection_wcs,
im.polarisation_frame)
return gcf_image, cf
def create_pswf_convolutionfunction(im, oversampling=8, support=6):
""" Fill an Anti-Aliasing filter into a ConvolutionFunction
Fill the Prolate Spheroidal Wave Function into a GriData with the specified oversampling. Only the inner
non-zero part is retained
Also returns the griddata correction function as an image
:param im: Image template
:param oversampling: Oversampling of the convolution function in uv space
:return: griddata correction Image, griddata kernel as ConvolutionFunction
"""
assert isinstance(im, Image)
# Calculate the convolution kernel. We oversample in u,v space by the factor oversampling
cf = create_convolutionfunction_from_image(im,
oversampling=oversampling,
support=support)
kernel = numpy.zeros([oversampling, support])
for grid in range(support):
for subsample in range(oversampling):
nu = ((grid - support // 2) -
(subsample - oversampling // 2) / oversampling)
kernel[subsample, grid] = grdsf([nu / (support // 2)])[1]
kernel /= numpy.sum(numpy.real(kernel[oversampling // 2, :]))
nchan, npol, _, _ = im.shape
cf.data = numpy.zeros(
[nchan, npol, 1, oversampling, oversampling, support,
support]).astype('complex')
for y in range(oversampling):
for x in range(oversampling):
cf.data[:, :, 0, y,
x, :, :] = numpy.outer(kernel[y, :],
kernel[x, :])[numpy.newaxis,
numpy.newaxis, ...]
norm = numpy.sum(numpy.real(cf.data[0, 0, 0, 0, 0, :, :]))
cf.data /= norm
# Now calculate the griddata correction function as an image with the same coordinates as the image
# which is necessary so that the correction function can be applied directly to the image
nchan, npol, ny, nx = im.data.shape
nu = numpy.abs(2.0 * coordinates(nx))
gcf1d = grdsf(nu)[0]
gcf = numpy.outer(gcf1d, gcf1d)
gcf[gcf > 0.0] = gcf.max() / gcf[gcf > 0.0]
gcf_data = numpy.zeros_like(im.data)
gcf_data[...] = gcf[numpy.newaxis, numpy.newaxis, ...]
gcf_image = create_image_from_array(gcf_data, cf.projection_wcs,
im.polarisation_frame)
return gcf_image, cf
def test_coordinates(self):
for N in [4, 5, 6, 7, 8, 9, 1000, 1001, 1002, 1003]:
low, high = coordinateBounds(N)
c = coordinates(N)
cx, cy = coordinates2(N)
self.assertAlmostEqual(numpy.min(c), low)
self.assertAlmostEqual(numpy.max(c), high)
self.assertAlmostEqual(numpy.min(cx), low)
self.assertAlmostEqual(numpy.max(cx), high)
self.assertAlmostEqual(numpy.min(cy), low)
self.assertAlmostEqual(numpy.max(cy), high)
assert c[N // 2] == 0
assert (cx[N // 2, :] == 0).all()
assert (cy[:, N // 2] == 0).all()