def plotCloudImage(self):
"""Generate some additional information and plots about the cloud image, if desired."""
from pImagePlots import PImagePlots
import pylab
im = PImagePlots()
im.setImage(self.cloudimage)
im.showImage(copy=True)
im.hanningFilter()
im.calcAll()
im.showPsd2d()
im.showAcovf2d()
im.showAcovf1d()
im.showSf(linear=True)
#pylab.show()
return
# image intensity information (and certainly lose zeropoint) #im.invertAcovf2d(usePhasespec=False) #im.invertAcovf2d(useI=True) # Start here to invert from 1d PSD (and use phase info or not) #im.invertPsd1d(phasespec=im.phasespec) # Start here to invert from 2d PSD (useI = False then, and uses phase info - get perfect reconstruction) im.invertPsd2d(usePhasespec=False) #im.invertPsd2d(useI=True) # Start here to invert from FFT (useI = False, and will get perfect reconstruction). im.invertFft(useI=True) # Use im2 to recalculate 1d PSD/ACovF starting from the reconstructed image, without altering the original. im2 = PImagePlots() im2.setImage(im.imageI.real, copy=True) im2.calcAll(min_dr=1.0, min_npix=2) im2.plotMore() # Now start plotting things, in comparison. clims = im.showImage() #print clims im2.showImage(clims=clims) im2.showImage() im.showFft(clims=clims) im2.showFft(clims=clims) im.showPsd2d() im2.showPsd2d() im.showPhases() im2.showPhases()
kappa = 1.
# And also try to make an image with the same (final) pixel scale, but that will be created (first)
# larger and then scaled down, to avoid introducing artifacts from the ACovF1d being truncated.
final_imsize = 1500 # pixels desired in final image
final_rad_fov = 1.75*numpy.sqrt(2) # degrees
final_pixscale = 2*final_rad_fov / float(final_imsize)
# Start with larger image
pixscale = final_pixscale
rad_fov = final_rad_fov
imsize = int(2*rad_fov / pixscale)
if (imsize%2 != 0):
imsize = imsize + 1
xr = xsf / pixscale # xr = in pixels, over range that want to simulate
print 'Image: Rad_fov', rad_fov, 'Imsize', imsize, 'Pixscale', pixscale, 'deg/pix', '(', pixscale*60.*60., 'arcsec/pix)'
im = PImagePlots(shift=True, nx=imsize, ny=imsize)
im.makeImageFromSf(sfx=xr, sf=SF)
im.showAcovf1d()
im.showPsd2dI()
# Trim image to desired final size
trim = round((imsize - final_imsize)/2.0)
print 'Trimming about %d pixels from each side' %(trim)
image = im.imageI[trim:trim+final_imsize, trim:trim+final_imsize]
image = rescaleImage(image.real, sigma_goal, kappa)
imsize = len(image)
pixscale = pixscale
rad_fov = imsize/2.0*pixscale
print 'Image After Trimming: Rad_fov', rad_fov, 'Imsize', imsize, 'Pixscale', pixscale, 'deg/pix', '(', pixscale*60.*60., 'arcsec/pix)'
im.setImage(image.real)
imageStats(im.image)
im.showImage(copy=True)
pixscale = 2*rad_fov / float(imsize)
print 'Rad_fov', rad_fov, 'Imsize', imsize, 'Pixscale', pixscale, 'deg/pix', '(', pixscale*60.*60., 'arcsec/pix)'
# set up 1d SF with desired scale and pixel scale
"""
condition = (xsf <= rad_fov)
xr = xsf[condition] / pixscale # xr = in pixels, over range that want to simulate
xrpix = numpy.arange(0, imsize, 1.0)
sfpix = numpy.interp(xrpix, xr, SF[condition])
"""
condition = (xsf <= rad_fov*numpy.sqrt(2))
xr = xsf[condition] / pixscale # xr = in pixels, over range that want to simulate
xrpix = numpy.arange(0, imsize/2.*numpy.sqrt(2), 1.0)
sfpix = numpy.interp(xrpix, xr, SF[condition])
# try making image
im = PImagePlots(shift=True, nx= imsize, ny=imsize)
im.makeImageFromSf(sfx=xrpix, sf=sfpix)
im.showImageI()
pylab.savefig('clouds_1.png', format='png')
im.image = im.imageI.real
#im.hanningFilter()
im.calcAll(min_npix=2, min_dr=1)
im.plotMore()
pylab.savefig('clouds_1_dat.png', format='png')
# Rescale sfx to be in physical (degrees)
im.sfx = im.sfx * pixscale
# make another image, as should see difference due to different random phases
im2 = PImagePlots(shift=True, nx= imsize, ny=imsize)
im2.invertSf(sfx=xrpix, sf=sfpix)
def clouds():
"""Read an example of the french group's cloud generation."""
# oldCloud.npy and newCloud.npy are images of size 240x240 that cover a fov of 4.0 deg
# (if cloud generation code is understood correctly).
# old clouds
oldClouds = numpy.load('oldCloud.npy')
fov = 4.0 #rad_fov = 2.0
nx = len(oldClouds)
pixscale = fov / float(nx)
im = PImagePlots(shift=True)
im.setImage(oldClouds)
im.showImage()
pylab.savefig('clouds_oldimage.%s' %(figformat), format='%s' %(figformat))
#im.hanningFilter()
im.calcAll(min_npix=2, min_dr=1)
im.plotMore()
pylab.savefig('clouds_old.%s' %(figformat), format='%s' %(figformat))
# new clouds
newClouds = numpy.load('newCloud.npy')
im2 = PImagePlots(shift=True)
im2.setImage(newClouds)
im2.showImage()
pylab.savefig('clouds_newimage.%s' %(figformat), format='%s' %(figformat))
#im2.hanningFilter()
im2.calcAll(min_npix=2, min_dr=1)
im2.plotMore()
pylab.savefig('clouds_new.%s' %(figformat), format='%s' %(figformat))
# compare structure functions
# translate x axis from pixels to degrees .. 240 pix = 4.0 deg (?)
im.sfx = im.sfx *pixscale
im2.sfx = im2.sfx *pixscale
# and scale SF's to just run between 0 and 1 (because of loss of amplitude info with random phases)
im.sf = im.sf / im.sf.max()
im2.sf = im2.sf / im2.sf.max()
legendlabels = ['Old clouds (scaled SF)', 'New clouds (scaled SF)']
im.showSf(comparison=im2, legendlabels=legendlabels, linear=True)
pylab.xlim(0, fov/2.0)
pylab.ylim(0, 1.2)
pylab.title('Structure Function')
pylab.xlabel('Degrees')
pylab.savefig('clouds_sf.%s' %(figformat), format='%s' %(figformat))
# look at phase spectrum
pylab.figure()
n, b, p = pylab.hist(im.phasespec.flatten(), bins=75, range=[-numpy.pi, numpy.pi],
alpha=0.2, label='Old clouds phases')
n, b, p = pylab.hist(im2.phasespec.flatten(), bins=b, range=[-numpy.pi, numpy.pi],
alpha=0.2, label='New clouds phases')
pylab.legend(fancybox=True, fontsize='smaller')
pylab.savefig('clouds_phasehist.%s' %(figformat), format='%s' %(figformat))
# the phase spectrum seems to be flatly distributed between -pi and pi
pylab.figure()
pylab.subplot(121)
pylab.title('Old clouds')
pylab.imshow(im.phasespec, origin='lower')
pylab.colorbar(shrink=0.6)
pylab.subplot(122)
pylab.title('New clouds')
pylab.imshow(im2.phasespec, origin='lower')
pylab.colorbar(shrink=0.6)
pylab.suptitle('Phase spectrum')
pylab.savefig('clouds_phasespec.%s' %(figformat), format='%s' %(figformat))
pylab.close()
return
def inversion():
"""Generate some example images & invert them to reconstruct the original image."""
im = TestImage(shift=True, nx=1000, ny=1000)
#im.addEllipseGrid(gridX=200, gridY=100, semiX=50, semiY=25, value=1)
im.addLines(width=20, spacing=200, value=1, angle=45)
im.addSin(scale=300)
im.hanningFilter()
im.zeroPad()
#cmap = pylab.cm.gray_r
cmap = None
clims = im.showImage(cmap=cmap)
pylab.savefig('invert_image.%s' %(figformat), format='%s' %(figformat))
im.calcAll(min_npix=1, min_dr=1)
# Invert from ACovF and show perfect reconstruction.
im.invertAcovf2d()
im.invertPsd2d(useI=True)
im.invertFft(useI=True)
im.showImageI(clims=clims, cmap=cmap)
pylab.savefig('invert_acovf2d_good.%s' %(figformat), format='%s' %(figformat))
# Invert from ACovF 2d without phases
im.invertAcovf2d(usePhasespec=False, seed=42)
im.invertPsd2d(useI=True)
im.invertFft(useI=True)
im.showImageI(clims=clims, cmap=cmap)
pylab.savefig('invert_acovf2d_nophases.%s' %(figformat), format='%s' %(figformat))
# Invert from ACovF 1d with phases
im.invertAcovf1d(phasespec=im.phasespec)
im.invertAcovf2d(useI=True)
im.invertPsd2d(useI=True)
im.invertFft(useI=True)
im.showImageI(clims=clims, cmap=cmap)
pylab.savefig('invert_acovf1d_phases.%s' %(figformat), format='%s' %(figformat))
# Invert from ACovF 1d without phases
im.invertAcovf1d(seed=42)
im.invertAcovf2d(useI=True)
im.invertPsd2d(useI=True)
im.invertFft(useI=True)
im.showImageI(clims=clims, cmap=cmap)
pylab.savefig('invert_acovf1d_nophases.%s' %(figformat), format='%s' %(figformat))
# Recalculate 1-d PSD and ACovF from this last reconstructed image (ACovF1d no phases)
im2 = PImagePlots()
im2.setImage(im.imageI)
im2.calcAll(min_npix=1, min_dr=1)
legendlabels=['Reconstructed', 'Original']
im2.showPsd1d(comparison=im, legendlabels=legendlabels)
pylab.savefig('invert_recalc_ACovF_Psd1d.%s' %(figformat), format='%s' %(figformat))
im2.showAcovf1d(comparison=im, legendlabels=legendlabels)
pylab.savefig('invert_recalc_ACovF_Acovf1d.%s' %(figformat), format='%s' %(figformat))
# Invert from PSD and show perfect reconstruction.
im.invertPsd2d()
im.invertFft(useI=True)
im.showImageI(clims=clims, cmap=cmap)
pylab.savefig('invert_psd2d_good.%s' %(figformat), format='%s' %(figformat))
# Invert from PSD 2d without phases
im.invertPsd2d(usePhasespec=False, seed=42)
im.invertFft(useI=True)
im.showImageI(clims=clims, cmap=cmap)
pylab.savefig('invert_psd2d_nophases.%s' %(figformat), format='%s' %(figformat))
# Invert from PSD 1d with phases
im.invertPsd1d(phasespec=im.phasespec)
im.invertPsd2d(useI=True)
im.invertFft(useI=True)
im.showImageI(clims=clims, cmap=cmap)
pylab.savefig('invert_psd1d_phases.%s' %(figformat), format='%s' %(figformat))
# Invert from PSD 1d without phases
im.invertPsd1d(seed=42)
im.invertPsd2d(useI=True)
im.invertFft(useI=True)
im.showImageI(clims=clims, cmap=cmap)
pylab.savefig('invert_psd1d_nophases.%s' %(figformat), format='%s' %(figformat))
# Recalculate 1-d PSD and ACovF from this last reconstructed image (PSD1d no phases)
im2 = PImagePlots()
im2.setImage(im.imageI)
im2.calcAll(min_npix=1, min_dr=1)
im2.showPsd1d(comparison=im, legendlabels=legendlabels)
pylab.savefig('invert_recalc_PSD_Psd1d.%s' %(figformat), format='%s' %(figformat))
im2.showAcovf1d(comparison=im, legendlabels=legendlabels)
pylab.savefig('invert_recalc_PSD_Acovf1d.%s' %(figformat), format='%s' %(figformat))
pylab.close()
return
