"""
This program convolves/smooths an image from a high resolution to a lower
resolution. This program works similar to a radio astronomy program where
we do convolution by cutting out the higher frequency components of the UV
plane, so point soutces appear broader but retain the same 'peak' response.
It uses the astropy convolution module
"""
import math
import sys
import timeit
from datetime import date
import astropy.visualization as vis
import matplotlib.pyplot as plt
import numpy as np
from astropy.convolution import (
Gaussian2DKernel,
Tophat2DKernel,
convolve,
convolve_fft,
)
from astropy.io import fits
from astropy.modeling.models import Gaussian2D
from astropy.wcs import WCS
from tw_source_finder.check_array import check_array, update_dimensions
[docs]def convolve_image(fits_input_image, conv_factor, use_fft="F", do_downsize="F"):
# Load the image to be convolved
print("loading input_image", fits_input_image)
hdu_list = fits.open(fits_input_image)
hdu = hdu_list[0]
incoming_dimensions = hdu.header["NAXIS"]
# get the pixel size - square images assumed
pixel_size = hdu.header["CDELT2"] * 3600
try:
bmaj_old = hdu.header["BMAJ"] * 3600
bmin_old = hdu.header["BMIN"] * 3600
bpa_old = hdu.header["BPA"]
bmaj = bmaj_old * conv_factor
bmin = bmin_old * conv_factor
bpa = bpa_old
beam_size_gain = (bmaj * bmin) / (bmaj_old * bmin_old)
if conv_factor > 1:
x_conv = math.sqrt(bmaj * bmaj - bmaj_old * bmaj_old)
y_conv = math.sqrt(bmin * bmin - bmin_old * bmin_old)
# NOTE: if conv_factor == 1, we still do a convolution,
# BUT the system will be convolving with respect to individual pixels,
# so can be used to 'smooth' simulated noise
else:
x_conv = bmaj
y_conv = bmin
# get reference position
w = WCS(hdu.header)
w = w.celestial
ref_ra = hdu.header["CRVAL1"]
ref_dec = hdu.header["CRVAL2"]
# determine ra, and dec of new reference pixel, which will be the midpoint
# of output image
cen_pos_x = hdu.header["NAXIS1"] // 2
cen_pos_y = hdu.header["NAXIS2"] // 2
lon, lat = w.all_pix2world(cen_pos_x, cen_pos_y, 0)
except:
print("Sorry - your input image must contain the FITS kewords BMAJ, BIN, and BPA")
return
# need to adjust for new beam vs old beam size ratio
x_stddev = x_conv / (2.355 * pixel_size)
y_stddev = y_conv / (2.355 * pixel_size)
theta = math.radians(bpa + 90.0) # theta from E-W line
kernel = Gaussian2DKernel(x_stddev, y_stddev, theta)
original_array = hdu.data
num_axes = len(hdu.data.shape)
# now convolved the image
img_source = check_array(hdu.data)
# set NaNs to zero
img_source = np.nan_to_num(img_source)
if use_fft == "T":
print("**** using FFT for convolution")
astropy_conv = convolve_fft(img_source, kernel, allow_huge=True)
else:
print("**** using image plane convolution")
astropy_conv = convolve(img_source, kernel)
# apply gain factor
astropy_conv = astropy_conv * beam_size_gain
# Now we plot te orginal and convolved image.
# plt.figure(1, figsize=(12, 12)).clf()
ax1 = plt.subplot(1, 2, 1, projection=WCS(hdu.header).celestial)
interval = vis.PercentileInterval(99.9)
vmin, vmax = interval.get_limits(img_source)
norm = vis.ImageNormalize(vmin=vmin, vmax=vmax, stretch=vis.LogStretch(1000))
im = ax1.imshow(img_source, cmap=plt.cm.gray_r, norm=norm, origin="lower")
ax1.coords["ra"].set_axislabel("Right Ascension")
ax1.coords["dec"].set_axislabel("Declination")
ax1.set_title("Original Image")
# plt.colorbar(im)
ax4 = plt.subplot(1, 2, 2, projection=WCS(hdu.header).celestial)
interval = vis.PercentileInterval(99.9)
vmin, vmax = interval.get_limits(astropy_conv)
norm = vis.ImageNormalize(vmin=vmin, vmax=vmax, stretch=vis.LogStretch(1000))
im = ax4.imshow(astropy_conv, cmap=plt.cm.gray_r, norm=norm, origin="lower")
ax4.coords["ra"].set_axislabel("Right Ascension")
ax4.coords["dec"].set_axislabel("Declination")
ax4.set_title("Convolved Image")
ax4.set_xticklabels([])
ax4.set_yticklabels([])
end_point = fits_input_image.find(".fits")
if end_point > -1:
plt.suptitle(fits_input_image[:end_point] + " convolved")
else:
plt.suptitle(fits_input_image + " convolved")
shape = astropy_conv.shape
conv_factor_int = int(conv_factor)
# see https://moonbooks.org/Articles/How-to-downsampling-a-matrix-by-averaging-elements-nn-with-numpy-in-python-/
if conv_factor_int > 1:
if end_point > -1:
outfile = fits_input_image[:end_point] + "_conv.fits"
else:
outfile = fits_input_image + "_conv.fits"
if do_downsize == "T":
print("shrinking convolved image")
smaller_astropy_conv = astropy_conv[::conv_factor_int, ::conv_factor_int]
else:
smaller_astropy_conv = astropy_conv
else:
outfile = fits_input_image
smaller_astropy_conv = astropy_conv
shape = smaller_astropy_conv.shape
shape_x = shape[0] // 2
shape_y = shape[1] // 2
# hdu.data = flush_fits(smaller_astropy_conv,hdu_list)
output_image = update_dimensions(smaller_astropy_conv, incoming_dimensions)
hdu.data = output_image
hdu.header["BMAJ"] = bmaj / 3600
hdu.header["BMIN"] = bmin / 3600
hdu.header["BPA"] = bpa
hdu.header["DATAMAX"] = hdu.data.max()
hdu.header["DATAMIN"] = hdu.data.min()
if do_downsize == "T":
hdu.header["CDELT1"] = hdu.header["CDELT1"] * conv_factor_int
hdu.header["CDELT2"] = hdu.header["CDELT2"] * conv_factor_int
# need to flip array references vs what's seen on the display
hdu.header["CRPIX1"] = int(shape_y)
hdu.header["CRPIX2"] = int(shape_x)
# no idea why I have to explicity wrap a float inside a float here
hdu.header["CRVAL1"] = float(lon)
hdu.header["CRVAL2"] = float(lat)
hdu.header.set("CONVFACT", conv_factor, "factor used to convolve input image")
today = date.today()
d4 = today.strftime("%b-%d-%Y")
hdu.header["HISTORY"] = d4 + " convolved by a factor " + str(conv_factor)
hdu.writeto(outfile, overwrite=True)
# We can examine the two images (this makes use of the wcsaxes package behind the scenes):
plt.savefig(fits_input_image[:end_point] + "_conv.png")
# plt.show()
[docs]def main(argv):
start_time = timeit.default_timer()
# argv[1] = incoming fits image
# argv[2] convolution factor
print("convolving image ", argv[1])
conv_factor = float(argv[2])
try:
use_fft = argv[3]
except:
use_fft = "F"
try:
do_downsize = argv[4]
except:
do_downsize = "T"
print("conv_image incoming image", argv[1])
print("calling convolve with do_downsize", do_downsize)
convolve_image(argv[1], conv_factor, use_fft, do_downsize)
elapsed = timeit.default_timer() - start_time
print("Run Time:", elapsed, "seconds")
if __name__ == "__main__":
main(sys.argv)