images.imfilter

The images.imfilter IRAF package provides an assortment of image filtering and convolution tasks.

Notes

For questions or comments please see our github page. We encourage and appreciate user feedback.

Python replacements for the images.imfilter tasks can be found in the Astropy and Scipy packages. Astropy convolution offers two convolution options, convolve() is better for small kernels, and convolve_fft() is better for larger kernels, please see the Astropy convolution doc page and Astropy Convolution How to for more details. For this notebook, we will use convolve. Check out the list of kernels and filters avaialble for Astropy, and Scipy

Although astropy.convolution is built on scipy, it offers several advantages: * can handle NaN values * improved options for boundaries * provided built in kernels

So when possible, we will be using astropy.convolution functions in this notebook.

You can select from the following boundary rules in astropy.convolution: * none * fill * wrap * extend

You can select from the following boundary rules in scipy.ndimage.convolution: * reflect * constant * nearest * mirror * wrap

Below we change the matplotlib colormap to viridis. This is temporarily changing the colormap setting in the matplotlib rc file.

Contents:

# Temporarily change default colormap to viridis
import matplotlib.pyplot as plt
plt.rcParams['image.cmap'] = 'viridis'

boxcar

Please review the Notes section above before running any examples in this notebook

The boxcar convolution does a boxcar smoothing with a given box size, and applies this running average to an array. Here we show a 2-D example using Box2DKernel, which is convinient for square box sizes.

# Standard Imports
import numpy as np

# Astronomy Specific Imports
from astropy.io import fits
from astropy.convolution import convolve as ap_convolve
from astropy.convolution import Box2DKernel

# Plotting Imports/Setup
import matplotlib.pyplot as plt
%matplotlib inline

# grab subsection of fits images
test_data = '/eng/ssb/iraf_transition/test_data/jczgx1ppq_flc.fits'
sci1 = fits.getdata(test_data,ext=1)
my_arr = sci1[700:1030,2250:2800]

# setup our kernel
box_kernel = Box2DKernel(3)
# perform convolution
result = ap_convolve(my_arr, box_kernel, normalize_kernel=True)

plt.imshow(box_kernel, interpolation='none', origin='lower')
plt.title('Kernel')
plt.colorbar()
plt.show()
_images/images.imfilter_11_0.png 
fig, axes = plt.subplots(nrows=1, ncols=2)
pmin,pmax = 10, 200
a = axes[0].imshow(my_arr,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[0].set_title('Before Convolution')
a = axes[1].imshow(result,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[1].set_title('After Convolution')

fig.subplots_adjust(right = 0.8,left=0)
cbar_ax = fig.add_axes([0.85, 0.15, 0.05, 0.7])
fig.colorbar(a, cax=cbar_ax)
plt.show()
_images/images.imfilter_12_0.png

convolve

Please review the Notes section above before running any examples in this notebook

The convolve task allows you to convolve your data array with a kernel of your own creation. Here we show a simple example of a rectangular kernel applied to a 10 by 10 array using the astropy.convolution.convolve function

# Standard Imports
import numpy as np

# Astronomy Specific Imports
from astropy.io import fits
from astropy.convolution import convolve as ap_convolve

# Plotting Imports/Setup
import matplotlib.pyplot as plt
%matplotlib inline

# grab subsection of fits images
test_data = '/eng/ssb/iraf_transition/test_data/jczgx1ppq_flc.fits'
sci1 = fits.getdata(test_data,ext=1)
my_arr = sci1[840:950,2350:2500]

# add nan's to test array
my_arr[40:50,60:70] = np.nan
my_arr[70:73,110:113] = np.nan

# setup our custom kernel
my_kernel = [[0,1,0],[1,0,1],[0,1,0],[1,0,1],[0,1,0]]
# perform convolution
result = ap_convolve(my_arr, my_kernel, normalize_kernel=True, boundary='wrap')

plt.imshow(my_kernel, interpolation='none', origin='lower')
plt.title('Kernel')
plt.colorbar()
plt.show()
_images/images.imfilter_18_0.png 
fig, axes = plt.subplots(nrows=1, ncols=2)
pmin,pmax = 10, 200
a = axes[0].imshow(my_arr,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[0].set_title('Before Convolution')
a = axes[1].imshow(result,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[1].set_title('After Convolution')

fig.subplots_adjust(right = 0.8,left=0)
cbar_ax = fig.add_axes([0.85, 0.15, 0.05, 0.7])
fig.colorbar(a, cax=cbar_ax)
plt.show()
_images/images.imfilter_19_0.png

Here is an example using masking with scipy.convolve

# Standard Imports
import numpy as np
from scipy.ndimage import convolve as sp_convolve

# Astronomy Specific Imports
from astropy.io import fits

# Plotting Imports/Setup
import matplotlib.pyplot as plt
%matplotlib inline

# grab subsection of fits images
test_data = '/eng/ssb/iraf_transition/test_data/jczgx1ppq_flc.fits'
sci1 = fits.getdata(test_data,ext=1)
my_arr = sci1[700:1030,2250:2800]

# setup our custom kernel
my_kernel = np.array([[0,1,0],[1,0,1],[0,1,0],[1,0,1],[0,1,0]]) * (1/7.0)
# perform convolution
result = sp_convolve(my_arr, my_kernel, mode='wrap')

plt.imshow(my_kernel, interpolation='none', origin='lower')
plt.title('Kernel')
plt.colorbar()
plt.show()
_images/images.imfilter_23_0.png 
fig, axes = plt.subplots(nrows=1, ncols=2)
pmin,pmax = 10, 200
a = axes[0].imshow(my_arr,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[0].set_title('Before Convolution')
a = axes[1].imshow(result,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[1].set_title('After Convolution')

fig.subplots_adjust(right = 0.8,left=0)
cbar_ax = fig.add_axes([0.85, 0.15, 0.05, 0.7])
fig.colorbar(a, cax=cbar_ax)
plt.show()
_images/images.imfilter_24_0.png

gauss

Please review the Notes section above before running any examples in this notebook

The gaussian kernel convolution applies a gaussian function convolution to your data array. The Gaussian2DKernel size is defined slightly differently from the IRAF version.

# Standard Imports
import numpy as np

# Astronomy Specific Imports
from astropy.io import fits
from astropy.convolution import convolve as ap_convolve
from astropy.convolution import Gaussian2DKernel

# Plotting Imports/Setup
import matplotlib.pyplot as plt
%matplotlib inline

# grab subsection of fits images
test_data = '/eng/ssb/iraf_transition/test_data/jczgx1ppq_flc.fits'
sci1 = fits.getdata(test_data,ext=1)
my_arr = sci1[700:1030,2250:2800]

# setup our kernel, with 6 sigma and a 3 in x by 5 in y size
gauss_kernel = Gaussian2DKernel(6, x_size=5, y_size=7)
# perform convolution
result = ap_convolve(my_arr, gauss_kernel, normalize_kernel=True)

gauss_kernel

<astropy.convolution.kernels.Gaussian2DKernel at 0x11ffc2350>

plt.imshow(gauss_kernel, interpolation='none', origin='lower')
plt.title('Kernel')
plt.colorbar()
plt.show()
_images/images.imfilter_30_0.png 
fig, axes = plt.subplots(nrows=1, ncols=2)
pmin,pmax = 10, 200
a = axes[0].imshow(my_arr,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[0].set_title('Before Convolution')
a = axes[1].imshow(result,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[1].set_title('After Convolution')

fig.subplots_adjust(right = 0.8,left=0)
cbar_ax = fig.add_axes([0.85, 0.15, 0.05, 0.7])
fig.colorbar(a, cax=cbar_ax)
plt.show()
_images/images.imfilter_31_0.png

laplace

Please review the Notes section above before running any examples in this notebook

The laplace task runs a image convolution using a laplacian filter with a subset of footprints. For the scipy.ndimage.filter.laplace function we will be using, you can feed any footprint in as an array to create your kernel.

# Standard Imports
import numpy as np
from scipy.ndimage import convolve as sp_convolve
from scipy.ndimage import laplace

# Astronomy Specific Imports
from astropy.io import fits

# Plotting Imports/Setup
import matplotlib.pyplot as plt
%matplotlib inline

# grab subsection of fits images
test_data = '/eng/ssb/iraf_transition/test_data/jczgx1ppq_flc.fits'
sci1 = fits.getdata(test_data,ext=1)
my_arr = sci1[700:1030,2250:2800]

# setup our laplace kernel with a target footprint (diagonals in IRAF)
footprint = np.array([[0, 1, 0], [1, 1, 1], [0, 1, 0]])
laplace_kernel = laplace(footprint)
# perform scipy convolution
result = sp_convolve(my_arr, laplace_kernel)

plt.imshow(laplace_kernel, interpolation='none', origin='lower')
plt.title('Kernel')
plt.colorbar()
plt.show()
_images/images.imfilter_37_0.png 
fig, axes = plt.subplots(nrows=1, ncols=2)
a = axes[0].imshow(my_arr,interpolation='none', origin='lower',vmin=0, vmax=70)
axes[0].set_title('Before Convolution')
a = axes[1].imshow(result,interpolation='none', origin='lower',vmin=0, vmax=70)
axes[1].set_title('After Convolution')

fig.subplots_adjust(right = 0.8,left=0)
cbar_ax = fig.add_axes([0.85, 0.15, 0.05, 0.7])
fig.colorbar(a, cax=cbar_ax)
plt.show()
_images/images.imfilter_38_0.png

median-rmedian

Please review the Notes section above before running any examples in this notebook

Apply a median filter to your data array. We will use the scipy.ndimage.filters.median_filter function.

# Standard Imports
import numpy as np
from scipy.ndimage.filters import median_filter

# Astronomy Specific Imports
from astropy.io import fits

# Plotting Imports/Setup
import matplotlib.pyplot as plt
%matplotlib inline

# create test array
test_data = '/eng/ssb/iraf_transition/test_data/jczgx1ppq_flc.fits'
sci1 = fits.getdata(test_data,ext=1)
my_arr = sci1[700:1030,2250:2800]

# apply median filter
filtered = median_filter(my_arr,size=(3,4))

fig, axes = plt.subplots(nrows=1, ncols=2)
pmin,pmax = 10, 200
a = axes[0].imshow(my_arr,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[0].set_title('Before Filter')
a = axes[1].imshow(filtered,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[1].set_title('After Filter')

fig.subplots_adjust(right = 0.8,left=0)
cbar_ax = fig.add_axes([0.85, 0.15, 0.05, 0.7])
fig.colorbar(a, cax=cbar_ax)
plt.show()
_images/images.imfilter_44_0.png

For a ring median filter we can supply a more specific footprint to the median_filter function. You can easily generate this footprint using the astroimtools library

# Standard Imports
import numpy as np
from scipy.ndimage.filters import median_filter

# Astronomy Specific Imports
from astropy.io import fits
from astroimtools import circular_annulus_footprint

# Plotting Imports/Setup
import matplotlib.pyplot as plt
%matplotlib inline

#depreciation warning, is fixed already in the dev version, not sure when this is getting pushed

# create test array
test_data = '/eng/ssb/iraf_transition/test_data/jczgx1ppq_flc.fits'
sci1 = fits.getdata(test_data,ext=1)
my_arr = sci1[700:1030,2250:2800]

# create annulus filter
fp = circular_annulus_footprint(10, 12)
# apply median filter
filtered = median_filter(my_arr, footprint=fp)

plt.imshow(fp, interpolation='none', origin='lower')
plt.title('Annulus Footprint')
plt.colorbar()
plt.show()
_images/images.imfilter_48_0.png 
fig, axes = plt.subplots(nrows=1, ncols=2)
pmin,pmax = 10, 200
a = axes[0].imshow(my_arr,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[0].set_title('Before Filter')
a = axes[1].imshow(filtered,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[1].set_title('After Filter')

fig.subplots_adjust(right = 0.8,left=0)
cbar_ax = fig.add_axes([0.85, 0.15, 0.05, 0.7])
fig.colorbar(a, cax=cbar_ax)
plt.show()
_images/images.imfilter_49_0.png

mode-rmode

Please review the Notes section above before running any examples in this notebook

The mode calculation equation used in the mode and rmode IRAF tasks (3.0*median - 2.0*mean) can be recreated using the scipy.ndimage.generic_filter function. The equation was used as an approximation for a mode calculation.

# Standard Imports
import numpy as np
from scipy.ndimage import generic_filter

# Astronomy Specific Imports
from astropy.io import fits

# Plotting Imports/Setup
import matplotlib.pyplot as plt
%matplotlib inline

def mode_func(in_arr):
    f = 3.0*np.median(in_arr) - 2.0*np.mean(in_arr)
    return f

For a box footprint:

# create test array
test_data = '/eng/ssb/iraf_transition/test_data/jczgx1ppq_flc.fits'
sci1 = fits.getdata(test_data,ext=1)
my_arr = sci1[700:1030,2250:2800]

# apply mode filter
filtered = generic_filter(my_arr,mode_func,size=5)

fig, axes = plt.subplots(nrows=1, ncols=2)
pmin,pmax = 10, 200
a = axes[0].imshow(my_arr,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[0].set_title('Before Filter')
a = axes[1].imshow(filtered,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[1].set_title('After Filter')

fig.subplots_adjust(right = 0.8,left=0)
cbar_ax = fig.add_axes([0.85, 0.15, 0.05, 0.7])
fig.colorbar(a, cax=cbar_ax)
plt.show()
_images/images.imfilter_57_0.png

For a ring footprint:

# Standard Imports
import numpy as np
from scipy.ndimage import generic_filter

# Astronomy Specific Imports
from astropy.io import fits
from astroimtools import circular_annulus_footprint

# Plotting Imports/Setup
import matplotlib.pyplot as plt
%matplotlib inline

# create test array
test_data = '/eng/ssb/iraf_transition/test_data/jczgx1ppq_flc.fits'
sci1 = fits.getdata(test_data,ext=1)
my_arr = sci1[700:1030,2250:2800]

# create annulus filter
fp = circular_annulus_footprint(5, 9)
# apply mode filter
filtered = generic_filter(my_arr,mode_func,footprint=fp)

plt.imshow(fp, interpolation='none', origin='lower')
plt.title('Annulus Footprint')
plt.colorbar()
plt.show()
_images/images.imfilter_61_0.png 
fig, axes = plt.subplots(nrows=1, ncols=2)
pmin,pmax = 10, 200
a = axes[0].imshow(my_arr,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[0].set_title('Before Filter')
a = axes[1].imshow(filtered,interpolation='none', origin='lower',vmin=pmin, vmax=pmax)
axes[1].set_title('After Filter')

fig.subplots_adjust(right = 0.8,left=0)
cbar_ax = fig.add_axes([0.85, 0.15, 0.05, 0.7])
fig.colorbar(a, cax=cbar_ax)
plt.show()
_images/images.imfilter_62_0.png

Not Replacing