def register_images(files, params):
"""Register the images in the given files list.
Registered images are placed in the loc_output dir with prefix r_
"""
# Have we specified a registration template?
if params.registration_image:
reg = DS.Observation(params.registration_image, params)
else:
reg = DS.EmptyBase()
reg.fw = 999.0
for f in files:
if (f.fw < reg.fw) and (f.fw > 1.2):
reg = f
print('Registration image:', reg.name)
# Register images
for f in files:
if f == reg:
f.image = f.data
rf = params.loc_output + os.path.sep + 'r_' + f.name
IO.write_image(f.image, rf)
else:
f.register(reg, params)
# delete image arrays to save memory
del f.image
del f.mask
del f.inv_variance
del reg.data
del reg.image
del reg.mask
del reg.inv_variance
return(files)
def imsub_all_fits(params, reference='ref.fits'):
#
# Create the output directory if it doesn't exist
#
if not (os.path.exists(params.loc_output)):
os.mkdir(params.loc_output)
#
# The degree of spatial shape changes has to be at least as
# high as the degree of spatial photometric scale
#
if (params.sdeg < params.pdeg):
print 'Increasing params.sdeg to ', params.pdeg
params.sdeg = params.pdeg
#
# Print out the parameters for this run.
#
print 'Parameters:'
for par in dir(params):
print par, getattr(params, par)
print
#
# Determine our list of images
#
all_files = os.listdir(params.loc_data)
all_files.sort()
files = []
for f in all_files:
print 'file', f
if fnmatch.fnmatch(f, params.name_pattern):
g = DS.Observation(params.loc_data + os.path.sep + f, params)
del g.data
del g.mask
print 'fw', g.fw
if g.fw > 0.0:
files.append(g)
print g.name, 'accepted'
if len(files) < 3:
print 'Only', len(files), 'files found matching', params.name_pattern
print 'Exiting'
sys.exit(0)
#
# Have we specified a registration template?
#
if params.registration_image:
reg = DS.Observation(params.registration_image, params)
else:
reg = DS.EmptyBase()
reg.fw = 999.0
for f in files:
if (f.fw < reg.fw) and (f.fw > params.reference_min_seeing) and (
f.sky < params.registration_max_background):
reg = f
print 'Registration image:', reg.name
#
# Register images
#
print 'Registering images'
files_copy = [f for f in files]
for f in files:
print f.name
if f == reg:
f.image = f.data
rf = params.loc_output + os.path.sep + 'r_' + f.name
IO.write_image(f.image, rf)
else:
if not f.register(reg, params):
files_copy.remove(f)
# delete image arrays to save memory
del f.image
del f.mask
del f.inv_variance
del reg.data
del reg.image
del reg.mask
del reg.inv_variance
files = files_copy
#
# Write image names and dates to a file
#
if params.image_list_file:
try:
with open(params.loc_output + os.path.sep + params.image_list_file,
'w') as fid:
for f in files:
date = None
if params.datekey:
date = IO.get_date(
params.loc_data + os.path.sep + f.name,
key=params.datekey) - 2450000
if date:
fid.write(f.name + ' %10.5f\n' % date)
else:
fid.write(f.name)
except:
raise
#
# Make the photometric reference image if we don't have it.
# Find stamp positions if required.
#
if not (os.path.exists(params.loc_output + os.path.sep + reference)):
print 'Reg = ', reg.name
stamp_positions = make_reference(files,
reg,
params,
reference_image=reference)
ref = DS.Observation(params.loc_output + os.path.sep + reference,
params)
mask, _ = IO.read_fits_file(params.loc_output + os.path.sep + 'mask_' +
reference)
variance, _ = IO.read_fits_file(params.loc_output + os.path.sep +
'var_' + reference)
ref.mask = mask
ref.inv_variance = 1.0 / variance
ref.register(reg, params)
else:
ref = DS.Observation(params.loc_output + os.path.sep + reference,
params)
if os.path.exists(params.loc_output + os.path.sep + 'mask_' +
reference):
mask, _ = IO.read_fits_file(params.loc_output + os.path.sep +
'mask_' + reference)
else:
mask = np.ones_like(ref.data)
ref.mask = mask
ref.register(reg, params)
stamp_positions = None
if params.use_stamps:
stamp_file = params.loc_output + os.path.sep + 'stamp_positions'
if os.path.exists(stamp_file):
stamp_positions = np.genfromtxt(stamp_file)
else:
stars = PF.choose_stamps(ref, params)
stamp_positions = stars[:, 0:2]
np.savetxt(stamp_file, stamp_positions)
pm = params.pixel_max
params.pixel_max *= 0.9
ref.mask *= IM.compute_saturated_pixel_mask(ref.image, params)
params.pixel_max = pm
ref.blur = IM.boxcar_blur(ref.image)
if params.mask_cluster:
ref.mask *= IM.mask_cluster(ref.image, ref.mask, params)
#
# Detect stars and compute the PSF if we are doing photometry
#
star_positions = None
sky = 0.0
if params.do_photometry:
star_file = params.loc_output + os.path.sep + 'star_positions'
psf_file = params.loc_output + os.path.sep + 'psf.fits'
if not (os.path.exists(psf_file)) or not (os.path.exists(star_file)):
stars = PH.compute_psf_image(params, ref, psf_image=psf_file)
star_positions = stars[:, 0:2]
star_sky = stars[:, 4]
if os.path.exists(star_file):
star_positions = np.genfromtxt(star_file)
star_sky = star_positions[:, 0] * 0.0
else:
np.savetxt(star_file, star_positions)
print 'sky =', sky
#
# If we have pre-determined star positions
#
if params.star_file:
stars = np.genfromtxt(params.star_file)
star_positions = stars[:, 1:3]
if params.star_reference_image:
star_ref, h = IO.read_fits_file(params.star_reference_image)
offset, _, _ = register_translation(star_ref, ref.image, 1000)
dy, dx = offset
#dy, dx = IM.positional_shift(ref.image,star_ref)
print 'position shift =', dx, dy
star_positions[:, 0] -= dx
star_positions[:, 1] -= dy
np.savetxt(star_file, star_positions)
#
# If we are using a CPU, group the stars by location
#
print 'Group_check'
print 'params.do_photometry', params.do_photometry
print 'params.use_GPU', params.use_GPU
if params.do_photometry:
star_group_boundaries = None
detector_mean_positions_x = None
detector_mean_positions_y = None
star_unsort_index = None
star_sort_index,star_group_boundaries,detector_mean_positions_x,detector_mean_positions_y = \
PH.group_stars_ccd(params,star_positions,params.loc_output+os.path.sep+reference)
star_positions = star_positions[star_sort_index]
star_sky = star_sky[star_sort_index]
star_unsort_index = np.argsort(star_sort_index)
#
# Do photometry of the reference image
#
if params.do_photometry:
ref_flux_file = params.loc_output + os.path.sep + 'ref.flux'
if not (os.path.exists(ref_flux_file)):
result = difference_image(
ref,
ref,
params,
stamp_positions=stamp_positions,
psf_image=psf_file,
star_positions=star_positions,
star_group_boundaries=star_group_boundaries,
detector_mean_positions_x=detector_mean_positions_x,
detector_mean_positions_y=detector_mean_positions_y,
star_sky=star_sky)
if isinstance(result.flux, np.ndarray):
print 'ungrouping fluxes'
result.flux = result.flux[star_unsort_index].copy()
result.dflux = result.dflux[star_unsort_index].copy()
np.savetxt(ref_flux_file,
np.vstack((result.flux, result.dflux)).T)
#
# Process images
#
if params.make_difference_images:
if not (params.use_GPU) and (params.n_parallel > 1):
pool = Pool(params.n_parallel)
pool.map(
process_image_helper,
itertools.izip(
files,
itertools.repeat(
(ref, params, stamp_positions, star_positions,
star_group_boundaries, star_unsort_index,
detector_mean_positions_x,
detector_mean_positions_y))))
else:
for f in files:
process_image(
f, (ref, params, stamp_positions, star_positions,
star_group_boundaries, star_unsort_index,
detector_mean_positions_x, detector_mean_positions_y))
return files
def difference_image(ref,
target,
params,
stamp_positions=None,
psf_image=None,
star_positions=None,
star_group_boundaries=None,
detector_mean_positions_x=None,
detector_mean_positions_y=None,
star_sky=None,
kernelRadius=None,
kernel_inner_rad=7):
from scipy.linalg import lu_solve, lu_factor, LinAlgError
start = time.time()
print 'difference_image', ref.name, target.name
#
# Set the kernel size based on the difference in seeing from the reference
#
#kernelRadius = min(params.kernel_maximum_radius,
# max(params.kernel_minimum_radius,
# np.abs(target.fw-ref.fw)*params.fwhm_mult))
if kernelRadius is None:
kernelRadius = min(
params.kernel_maximum_radius,
max(params.kernel_minimum_radius,
np.sqrt(np.abs(target.fw**2 - ref.fw**2)) * params.fwhm_mult))
#
# Mask saturated pixels
#
#print 'Masking ',target.name,time.time()-start
#smask = compute_saturated_pixel_mask(target.image,kernelRadius,params)
#
# Define the kernel basis functions
#
print 'Defining kernel pixels', time.time() - start
if params.use_fft_kernel_pixels:
kernelIndex, extendedBasis = IM.define_kernel_pixels_fft(
ref,
target,
kernelRadius + 2,
INNER_RADIUS=20,
threshold=params.fft_kernel_threshold)
else:
kernelIndex, extendedBasis = IM.define_kernel_pixels(
kernelRadius, INNER_RADIUS=kernel_inner_rad)
nKernel = kernelIndex.shape[0]
#
# We dont want to use bad pixels in either the target or reference image
#
smask = target.mask * ref.mask
bmask = np.ones(smask.shape, dtype=bool)
g = DS.EmptyBase()
for iteration in range(params.iterations):
print 'Computing matrix', time.time() - start
tmask = bmask * smask
#
# Compute the matrix and vector
#
H, V, texref = CI.compute_matrix_and_vector_cuda(
ref.image,
ref.blur,
target.image,
target.inv_variance,
tmask,
kernelIndex,
extendedBasis,
kernelRadius,
params,
stamp_positions=stamp_positions)
#
# Solve the matrix equation to find the kernel coefficients
#
print 'Solving matrix equation', time.time() - start
try:
lu, piv = lu_factor(H)
c = lu_solve((lu, piv), V).astype(np.float32).copy()
except (LinAlgError, ValueError):
print 'LU decomposition failed'
g.model = None
g.flux = None
g.diff = None
print 'H'
print H
sys.stdout.flush()
return g
#
# Compute the model image
#
print 'Computing model', time.time() - start
g.model = CI.compute_model_cuda(ref.image.shape, texref, c,
kernelIndex, extendedBasis, params)
edges = np.where(ref.image < 1.0)
g.model[edges] = 0.0
#
# Compute the difference image
#
difference = (target.image - g.model)
g.norm = difference * np.sqrt(target.inv_variance)
#
# Recompute the variance image from the model
#
#target.inv_variance = 1.0/(g.model/params.gain +
# (params.readnoise/params.gain)**2) + (1-smask)
mp = np.where(tmask == 0)
if len(mp[0]) > 0:
target.inv_variance[mp] = 1.e-12
#
# Mask pixels that disagree with the model
#
if iteration > 2:
bmask = IM.kappa_clip(smask, g.norm,
params.pixel_rejection_threshold)
print 'Iteration', iteration, 'completed', time.time() - start
#
# Delete the target image array to save memory
#
del target.image
#
# Save the kernel coefficients to a file
#
if params.do_photometry and psf_image:
kf = params.loc_output + os.path.sep + 'k_' + os.path.basename(
target.name)
IO.write_kernel_table(kf, kernelIndex, extendedBasis, c, params)
print 'coeffs', c
g.norm = difference * np.sqrt(target.inv_variance)
g.variance = 1.0 / target.inv_variance
g.mask = tmask
#
# Do the photometry if requested
#
g.flux = None
if params.do_photometry and psf_image:
print 'star_positions', star_positions.shape
print 'star_group_boundaries', star_group_boundaries
if ref.name == target.name:
sky_image, _ = IO.read_fits_file(params.loc_output + os.path.sep +
'temp.sub2.fits')
phot_target = ref.image - sky_image
IO.write_image(
phot_target,
params.loc_output + os.path.sep + 'clean_' + ref.name)
g.flux, g.dflux = CIF.photom_all_stars_simultaneous(
phot_target, target.inv_variance, star_positions, psf_image, c,
kernelIndex, extendedBasis, kernelRadius, params,
star_group_boundaries, detector_mean_positions_x,
detector_mean_positions_y)
else:
phot_target = difference
g.flux, g.dflux = CI.photom_all_stars(
phot_target, target.inv_variance, star_positions, psf_image, c,
kernelIndex, extendedBasis, kernelRadius, params,
star_group_boundaries, detector_mean_positions_x,
detector_mean_positions_y)
print 'Photometry completed', time.time() - start
#
# Apply the photometric scale factor to the difference image.
# We don't do this prior to the photometry because the PSF is
# being convolved by the kernel, which already includes the
# photometric scale factor.
#
g.diff = IM.apply_photometric_scale(difference, c, params.pdeg)
sys.stdout.flush()
return g