| 1 | package PDL::Demos::Transform_demo; | ||
| 2 | |||
| 3 | 0 | ▶ | use PDL::Graphics::Simple; |
|
Subroutine
BEGIN: called CC=1 SCAR=2.6
| |||
| 4 | use PDL::Transform; | ||
| 5 | require File::Spec; | ||
| 6 | use Carp; | ||
| 7 | |||
| 8 | sub info {('transform', 'Coordinate transformations (Req.: PDL::Graphics::Simple)')} | ||
| 9 | |||
| 10 | sub init {' | ||
| 11 | use PDL::Graphics::Simple; | ||
| 12 | '} | ||
| 13 | |||
| 14 | # try and find m51.fits | ||
| 15 | my @f = qw(PDL Demos m51.fits); | ||
| 16 | our $m51file = undef; | ||
| 17 | foreach my $path ( @INC ) { | ||
| 18 | my $file = File::Spec->catfile( $path, @f ); | ||
| 19 | if ( -f $file ) { $m51file = $file; last; } | ||
| 20 | } | ||
| 21 | confess "Unable to find m51.fits within the perl libraries.\n" | ||
| 22 | unless defined $m51file; | ||
| 23 | |||
| 24 | my @demo = ( | ||
| 25 | [comment => q| | ||
| 26 | This demo illustrates the PDL::Transform module. | ||
| 27 | |||
| 28 | It requires PDL::Graphics::Simple installed and makes use of the image of | ||
| 29 | M51 kindly provided by the Hubble Heritage group at the | ||
| 30 | Space Telescope Science Institute. | ||
| 31 | |||
| 32 | |], | ||
| 33 | |||
| 34 | [act => q| | ||
| 35 | # PDL::Transform objects embody coordinate transformations. | ||
| 36 | |||
| 37 | use PDL::Transform; | ||
| 38 | |||
| 39 | # set up a simple linear scale-and-shift relation | ||
| 40 | |||
| 41 | $t = t_linear( Scale=>[2,-1], Post=>[100,0]); | ||
| 42 | print $t; | ||
| 43 | |], | ||
| 44 | |||
| 45 | [act => q| | ||
| 46 | # The simplest way to use PDL::Transform is to transform a set of | ||
| 47 | # vectors. To do this you use the "apply" method. | ||
| 48 | |||
| 49 | # Define a few 2-vectors: | ||
| 50 | $xy = pdl([[0,1],[1,2],[10,3]]); | ||
| 51 | print "xy: ", $xy; | ||
| 52 | |||
| 53 | # Transform the 2-vectors: | ||
| 54 | print "Transformed: ", $xy->apply( $t ); | ||
| 55 | |], | ||
| 56 | |||
| 57 | [act => q| | ||
| 58 | # You can invert and compose transformations with 'x' and '!'. | ||
| 59 | $u = t_linear( Scale=>10 ); # A new transformation (simple x10 scale) | ||
| 60 | $xy = pdl([[0,1],[10,3]]); # Two 2-vectors | ||
| 61 | print "xy: ", $xy; | ||
| 62 | print "xy': ", $xy->apply( !$t ); # Invert $t from earlier. | ||
| 63 | print "xy'': ", $xy->apply( $u x !$t ); # Hit the result with $u. | ||
| 64 | |], | ||
| 65 | |||
| 66 | [act => q| | ||
| 67 | # PDL::Transform is useful for data resampling, and that's perhaps | ||
| 68 | # the best way to demonstrate it. First, we do a little bit of prep work: | ||
| 69 | |||
| 70 | # Read in an image ($m51file has been set up by this demo to | ||
| 71 | # contain the location of the file). Transform is designed to | ||
| 72 | # work well with FITS images that contain WCS scientific coordinate | ||
| 73 | # information, but works equally well in pixel space. | ||
| 74 | |||
| 75 | $m51 = rfits($|.__PACKAGE__.q|::m51file,{hdrcpy=>1}); | ||
| 76 | |||
| 77 | # we use a floating-point version of the image in some of the demos | ||
| 78 | # to highlight the interpolation schemes. (Note that the FITS | ||
| 79 | # header gets deep-copied automatically into the new variable). | ||
| 80 | |||
| 81 | $m51_fl = $m51->float; | ||
| 82 | |||
| 83 | # Define a nice, simple scale-by-3 transformation. | ||
| 84 | |||
| 85 | $ts = t_scale(3); | ||
| 86 | |], | ||
| 87 | |||
| 88 | [act => q| | ||
| 89 | #### Resampling with ->map and no FITS interpretation works in pixel space. | ||
| 90 | |||
| 91 | ### Create a plot window, and display the original image | ||
| 92 | $win = pgswin( size=>[8,6], multi=>[2,2] ) ; | ||
| 93 | $win->plot(with=>'image', $m51, { Title=>"M51" }); | ||
| 94 | |||
| 95 | ### Grow m51 by a factor of 3; origin is at lower left | ||
| 96 | # (the "pix" makes the resampling happen in pixel coordinate | ||
| 97 | # space, ignoring the FITS header) | ||
| 98 | |||
| 99 | $win->plot(with=>'image', $m51->map($ts, {pix=>1}), | ||
| 100 | { Title=>"M51 grown by 3 (pixel coords)" }); | ||
| 101 | |||
| 102 | ### Shrink m51 by a factor of 3; origin still at lower left. | ||
| 103 | # (You can invert the transform with a leading '!'.) | ||
| 104 | |||
| 105 | $win->plot(with=>'image', $m51->map(!$ts, {pix=>1}), | ||
| 106 | { Title=>"M51 shrunk by 3 (pixel coords)" }); | ||
| 107 | |], | ||
| 108 | |||
| 109 | [act => q| | ||
| 110 | # You can work in scientific space (or any other space) by | ||
| 111 | # wrapping your main transformation with something that translates | ||
| 112 | # between the coordinates you want to act in, and the coordinates | ||
| 113 | # you have. Here, "t_fits" translates between pixels in the data | ||
| 114 | # and arcminutes in the image plane. | ||
| 115 | |||
| 116 | $win->plot(with=>'points', pdl([1]), {title=>''}); # blank, clears | ||
| 117 | $win->plot(with=>'image', $m51, { Title=>"M51" }); | ||
| 118 | |||
| 119 | ### Scale in scientific coordinates. | ||
| 120 | # Here's a way to scale in scientific coordinates: | ||
| 121 | # wrap our transformation in FITS-header transforms to translate | ||
| 122 | # the transformation into scientific space. | ||
| 123 | |||
| 124 | $win->plot(with=>'image', $m51->map(!$ts->wrap(t_fits($m51)), {pix=>1}), | ||
| 125 | { Title=>"M51 shrunk by 3 (sci. coords)" }); | ||
| 126 | |], | ||
| 127 | |||
| 128 | [act => q| | ||
| 129 | # If you don't specify "pix=>1" then the resampler works in scientific | ||
| 130 | # FITS coordinates (if the image has a FITS header): | ||
| 131 | |||
| 132 | ### Scale in scientific coordinates (origin at center of galaxy) | ||
| 133 | $win->plot(with=>'fits', $m51->map($ts, $m51->hdr), { Title=>"M51 3x" }); | ||
| 134 | |||
| 135 | ### Instead of setting up a coordinate transformation you can use the | ||
| 136 | # implicit FITS header matching. Just tweak the template header: | ||
| 137 | $tohdr = $m51->hdr_copy; | ||
| 138 | $tohdr->{CDELT1} /= 3; # Magnify 3x in horiz direction | ||
| 139 | $tohdr->{CDELT2} /= 3; # Magnify 3x in vert direction | ||
| 140 | |||
| 141 | ### Resample to match the new FITS header | ||
| 142 | # (Note that, although the image is scaled exactly the same as before, | ||
| 143 | # this time the scientific coordinates have scaled too.) | ||
| 144 | $win->plot(with=>'fits', $m51->map(t_identity(), $tohdr), | ||
| 145 | { Title=>"3x (FITS)" }); | ||
| 146 | |], | ||
| 147 | |||
| 148 | [act => q| | ||
| 149 | ### The three main resampling methods are "sample", "linear", and "jacobian". | ||
| 150 | # Sampling is fastest, linear interpolation is better. Jacobian resampling | ||
| 151 | # is slow but prevents aliasing under skew or reducing transformations. | ||
| 152 | |||
| 153 | $win->plot(with=>'fits', $m51, { Title=>"M51" }); | ||
| 154 | |||
| 155 | $win->plot(with=>'fits', $m51_fl->map( $ts, $m51_fl, { method=>"sample" } ), | ||
| 156 | { Title=>"M51 x3 (sampled)" }); | ||
| 157 | |||
| 158 | $win->plot(with=>'fits', $m51_fl->map($ts, $m51_fl, {method=>"linear"}), | ||
| 159 | { Title=>"M51 x3 (interp.)"}); | ||
| 160 | |||
| 161 | $win->plot(with=>'fits', $m51_fl->map($ts, $m51_fl, { method=>"jacobian" }), | ||
| 162 | { Title=>"M51 x3 (jacob.)"}); | ||
| 163 | |], | ||
| 164 | |||
| 165 | [act => q| | ||
| 166 | ### Linear transformations are only the beginning. Here's an example | ||
| 167 | # using a simple nonlinear transformation: radial coordinate transformation. | ||
| 168 | |||
| 169 | ### Original image | ||
| 170 | $win->plot(with=>'fits', $m51, { Title=>"M51" }); | ||
| 171 | |||
| 172 | ### Radial structure in M51 (linear radial scale; origin at (0,0) by default) | ||
| 173 | $tu = t_radial( u=>'degree' ); | ||
| 174 | $win->plot(with=>'fits', $m51_fl->map($tu), | ||
| 175 | { Title=>"M51 radial (linear)", J=>0 }); | ||
| 176 | |||
| 177 | ### Radial structure in M51 (conformal/logarithmic radial scale) | ||
| 178 | $tu_c = t_radial( r0=>0.1 ); # Y axis 0 is at 0.1 arcmin | ||
| 179 | $win->plot(with=>'fits', $m51_fl->map($tu_c), | ||
| 180 | { Title=>"M51 radial (conformal)", YRange=>[0,4] } ); | ||
| 181 | |], | ||
| 182 | # NOTE: | ||
| 183 | # need to 'double protect' the \ in the label_axes() | ||
| 184 | # since it's being evaluated twice (I think) | ||
| 185 | # | ||
| 186 | |||
| 187 | [act => q| | ||
| 188 | |||
| 189 | ##################### | ||
| 190 | # Wrapping transformations allows you to work in a convenient | ||
| 191 | # space for what you want to do. Here, we can use a simple | ||
| 192 | # skew matrix to find (and remove) logarithmic spiral structures in | ||
| 193 | # the galaxy. The "unspiraled" images shift the spiral arms into | ||
| 194 | # approximate straight lines. | ||
| 195 | |||
| 196 | $sp = 3.14159; # Skew by 3.14159 | ||
| 197 | |||
| 198 | # Skew matrix | ||
| 199 | $t_skew = t_linear(pre => [$sp * 130, 0] , matrix => pdl([1,0],[-$sp,1])); | ||
| 200 | |||
| 201 | # When put into conformal radial space, the skew turns into 3.14159 | ||
| 202 | # radians per scale height. | ||
| 203 | $t_untwist = t_wrap($t_skew, $tu_c); | ||
| 204 | |||
| 205 | # Press enter to see the result of these transforms... | ||
| 206 | |], | ||
| 207 | |||
| 208 | [act => q| | ||
| 209 | ############################## | ||
| 210 | # Note that you can use ->map and ->unmap as either PDL methods | ||
| 211 | # or transform methods; what to do is clear from context. | ||
| 212 | |||
| 213 | $win->plot(with=>'points', pdl([1]), {title=>''}); # blank, clears | ||
| 214 | # Original image | ||
| 215 | $win->plot(with=>'fits', $m51, { Title=>"M51" }); | ||
| 216 | |||
| 217 | # Skewed | ||
| 218 | $win->plot(with=>'fits', $m51_fl->map( $t_skew ), | ||
| 219 | { Title => "M51 skewed by pi in spatial coords" } ); | ||
| 220 | |||
| 221 | # Untwisted -- show that m51 has a half-twist per scale height | ||
| 222 | $win->plot(with=>'fits', $m51_fl->map( $t_untwist ), | ||
| 223 | { Title => "M51 unspiraled (pi / r_s)"} ); | ||
| 224 | |||
| 225 | # Untwisted -- the jacobian method uses variable spatial filtering | ||
| 226 | # to eliminate spatial artifacts, at significant computational cost | ||
| 227 | # (This may take some time to complete). | ||
| 228 | $win->plot(with=>'fits', $m51_fl->map( $t_untwist, {m=>"jacobian"}), | ||
| 229 | { Title => "M51 unspiraled (pi / r_s; antialiased)" } ); | ||
| 230 | |], | ||
| 231 | |||
| 232 | [act => q| | ||
| 233 | ### Native FITS interpretation makes it easy to view your data in | ||
| 234 | ### your preferred coordinate system. Here we zoom in on a 0.2x0.2 | ||
| 235 | ### arcmin region of M51, sampling it to 100x100 pixels resolution. | ||
| 236 | |||
| 237 | $m51 = float $m51; | ||
| 238 | $data = $m51->match([100,100],{or=>[[-0.05,0.15],[-0.05,0.15]]}); | ||
| 239 | $s = "M51 closeup ("; $ss=" coords)"; | ||
| 240 | $ps = " (pixels)"; | ||
| 241 | |||
| 242 | $win = pgswin( size=>[8,4], multi=>[2,1] ) ; | ||
| 243 | $win->plot(with=>'image', $data, { title=>"${s}pixel${ss}", | ||
| 244 | xlabel=>"X$ps", ylabel=>"Y$ps", crange=>[600,750] } ); | ||
| 245 | $win->plot(with=>'fits', $data, { title=>"${s}sci.${ss}", | ||
| 246 | crange=>[600,750] } ); | ||
| 247 | |], | ||
| 248 | |||
| 249 | [act => q| | ||
| 250 | ### Now rotate the image 360 degrees in 10 degree increments. | ||
| 251 | ### The 'match' method resamples $data to the rotated scientific | ||
| 252 | ### coordinate system in $hdr. The "pixel coordinates" window shows | ||
| 253 | ### the resampled data in their new pixel coordinate system. | ||
| 254 | ### The "sci. coordinates" window shows the data remaining fixed in | ||
| 255 | ### scientific space, even though the pixels that represent them are | ||
| 256 | ### moving and rotating. | ||
| 257 | |||
| 258 | $hdr = $data->hdr_copy; | ||
| 259 | |||
| 260 | for ($rot=0; $rot<360; $rot += 10) { | ||
| 261 | $hdr->{CROTA2} = $rot; | ||
| 262 | |||
| 263 | $d = $m51->match($hdr); | ||
| 264 | |||
| 265 | $win->plot(with=>'image', $d, { title=>"${s}pixel${ss}", | ||
| 266 | xlabel=>"X$ps", ylabel=>"Y$ps", crange=>[600,750] } ); | ||
| 267 | $win->plot(with=>'fits', $d, { title=>"${s}sci.${ss}", | ||
| 268 | xrange=>[-0.05,0.15], yrange=>[-0.05,0.15], crange=>[600,750] } ); | ||
| 269 | } | ||
| 270 | |], | ||
| 271 | |||
| 272 | [act => q| | ||
| 273 | ### You can do the same thing even with nonsquare coordinates. | ||
| 274 | ### Here, we resample the same region in scientific space into a | ||
| 275 | ### 150x50 pixel array. | ||
| 276 | |||
| 277 | $data = $m51->match([150,50],{or=>[[-0.05,0.15],[-0.05,0.15]]}); | ||
| 278 | $hdr = $data->hdr_copy; | ||
| 279 | |||
| 280 | for ($rot=0; $rot<360; $rot += 5) { | ||
| 281 | $hdr->{CROTA2} = $rot; | ||
| 282 | $d = $m51->match($hdr,{or=>[[-0.05,0.15],[-0.05,0.15]]}); | ||
| 283 | $win->plot(with=>'image', $d, { title=>"${s}pixel${ss}", | ||
| 284 | xlabel=>"X$ps", ylabel=>"Y$ps", crange=>[600,750] } ); | ||
| 285 | $win->plot(with=>'fits', $d, { title=>"${s}sci.${ss}", | ||
| 286 | xrange=>[-0.05,0.15], yrange=>[-0.05,0.15], crange=>[600,750] } ); | ||
| 287 | } | ||
| 288 | |], | ||
| 289 | |||
| 290 | [comment => q| | ||
| 291 | This concludes the PDL::Transform demo. | ||
| 292 | |||
| 293 | Be sure to check the documentation for PDL::Transform::Cartography, | ||
| 294 | which contains common perspective and mapping coordinate systems | ||
| 295 | that are useful for work on the terrestrial and celestial spheres, | ||
| 296 | as well as other planets &c. | ||
| 297 | |], | ||
| 298 | ); | ||
| 299 | |||
| 300 | sub demo { @demo } | ||
| 301 | sub done {' | ||
| 302 | undef $win; | ||
| 303 | '} | ||
| 304 | |||
| 305 | 1; | ||