line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
|
2
|
|
|
|
|
|
|
# |
3
|
|
|
|
|
|
|
# GENERATED WITH PDL::PP! Don't modify! |
4
|
|
|
|
|
|
|
# |
5
|
|
|
|
|
|
|
package PDL::ImageND; |
6
|
|
|
|
|
|
|
|
7
|
|
|
|
|
|
|
@EXPORT_OK = qw( kernctr PDL::PP convolve ninterpol PDL::PP rebin circ_mean circ_mean_p PDL::PP convolveND ); |
8
|
|
|
|
|
|
|
%EXPORT_TAGS = (Func=>[@EXPORT_OK]); |
9
|
|
|
|
|
|
|
|
10
|
3
|
|
|
3
|
|
497
|
use PDL::Core; |
|
3
|
|
|
|
|
8
|
|
|
3
|
|
|
|
|
22
|
|
11
|
3
|
|
|
3
|
|
24
|
use PDL::Exporter; |
|
3
|
|
|
|
|
7
|
|
|
3
|
|
|
|
|
37
|
|
12
|
3
|
|
|
3
|
|
19
|
use DynaLoader; |
|
3
|
|
|
|
|
9
|
|
|
3
|
|
|
|
|
222
|
|
13
|
|
|
|
|
|
|
|
14
|
|
|
|
|
|
|
|
15
|
|
|
|
|
|
|
|
16
|
|
|
|
|
|
|
|
17
|
|
|
|
|
|
|
@ISA = ( 'PDL::Exporter','DynaLoader' ); |
18
|
|
|
|
|
|
|
push @PDL::Core::PP, __PACKAGE__; |
19
|
|
|
|
|
|
|
bootstrap PDL::ImageND ; |
20
|
|
|
|
|
|
|
|
21
|
|
|
|
|
|
|
|
22
|
|
|
|
|
|
|
|
23
|
|
|
|
|
|
|
|
24
|
|
|
|
|
|
|
|
25
|
|
|
|
|
|
|
=head1 NAME |
26
|
|
|
|
|
|
|
|
27
|
|
|
|
|
|
|
PDL::ImageND - useful image processing in N dimensions |
28
|
|
|
|
|
|
|
|
29
|
|
|
|
|
|
|
=head1 DESCRIPTION |
30
|
|
|
|
|
|
|
|
31
|
|
|
|
|
|
|
These routines act on PDLs as N-dimensional objects, not as threaded |
32
|
|
|
|
|
|
|
sets of 0-D or 1-D objects. The file is sort of a catch-all for |
33
|
|
|
|
|
|
|
broadly functional routines, most of which could legitimately |
34
|
|
|
|
|
|
|
be filed elsewhere (and probably will, one day). |
35
|
|
|
|
|
|
|
|
36
|
|
|
|
|
|
|
ImageND is not a part of the PDL core (v2.4) and hence must be explicitly |
37
|
|
|
|
|
|
|
loaded. |
38
|
|
|
|
|
|
|
|
39
|
|
|
|
|
|
|
=head1 SYNOPSIS |
40
|
|
|
|
|
|
|
|
41
|
|
|
|
|
|
|
use PDL::ImageND; |
42
|
|
|
|
|
|
|
|
43
|
|
|
|
|
|
|
$y = $x->convolveND($kernel,{bound=>'periodic'}); |
44
|
|
|
|
|
|
|
$y = $x->rebin(50,30,10); |
45
|
|
|
|
|
|
|
|
46
|
|
|
|
|
|
|
=cut |
47
|
|
|
|
|
|
|
|
48
|
|
|
|
|
|
|
|
49
|
|
|
|
|
|
|
|
50
|
|
|
|
|
|
|
|
51
|
|
|
|
|
|
|
|
52
|
|
|
|
|
|
|
|
53
|
|
|
|
|
|
|
|
54
|
|
|
|
|
|
|
|
55
|
|
|
|
|
|
|
=head1 FUNCTIONS |
56
|
|
|
|
|
|
|
|
57
|
|
|
|
|
|
|
|
58
|
|
|
|
|
|
|
|
59
|
|
|
|
|
|
|
=cut |
60
|
|
|
|
|
|
|
|
61
|
|
|
|
|
|
|
|
62
|
|
|
|
|
|
|
|
63
|
|
|
|
|
|
|
|
64
|
|
|
|
|
|
|
|
65
|
3
|
|
|
3
|
|
29
|
use Carp; |
|
3
|
|
|
|
|
6
|
|
|
3
|
|
|
|
|
954
|
|
66
|
|
|
|
|
|
|
|
67
|
|
|
|
|
|
|
|
68
|
|
|
|
|
|
|
|
69
|
|
|
|
|
|
|
|
70
|
|
|
|
|
|
|
|
71
|
|
|
|
|
|
|
=head2 convolve |
72
|
|
|
|
|
|
|
|
73
|
|
|
|
|
|
|
=for sig |
74
|
|
|
|
|
|
|
|
75
|
|
|
|
|
|
|
Signature: (a(m); b(n); indx adims(p); indx bdims(q); [o]c(m)) |
76
|
|
|
|
|
|
|
|
77
|
|
|
|
|
|
|
=for ref |
78
|
|
|
|
|
|
|
|
79
|
|
|
|
|
|
|
N-dimensional convolution (Deprecated; use convolveND) |
80
|
|
|
|
|
|
|
|
81
|
|
|
|
|
|
|
=for usage |
82
|
|
|
|
|
|
|
|
83
|
|
|
|
|
|
|
$new = convolve $x, $kernel |
84
|
|
|
|
|
|
|
|
85
|
|
|
|
|
|
|
Convolve an array with a kernel, both of which are N-dimensional. This |
86
|
|
|
|
|
|
|
routine does direct convolution (by copying) but uses quasi-periodic |
87
|
|
|
|
|
|
|
boundary conditions: each dim "wraps around" to the next higher row in |
88
|
|
|
|
|
|
|
the next dim. |
89
|
|
|
|
|
|
|
|
90
|
|
|
|
|
|
|
This routine is kept for backwards compatibility with earlier scripts; |
91
|
|
|
|
|
|
|
for most purposes you want L instead: |
92
|
|
|
|
|
|
|
it runs faster and handles a variety of boundary conditions. |
93
|
|
|
|
|
|
|
|
94
|
|
|
|
|
|
|
|
95
|
|
|
|
|
|
|
|
96
|
|
|
|
|
|
|
=for bad |
97
|
|
|
|
|
|
|
|
98
|
|
|
|
|
|
|
convolve does not process bad values. |
99
|
|
|
|
|
|
|
It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles. |
100
|
|
|
|
|
|
|
|
101
|
|
|
|
|
|
|
|
102
|
|
|
|
|
|
|
=cut |
103
|
|
|
|
|
|
|
|
104
|
|
|
|
|
|
|
|
105
|
|
|
|
|
|
|
|
106
|
|
|
|
|
|
|
|
107
|
|
|
|
|
|
|
|
108
|
|
|
|
|
|
|
|
109
|
|
|
|
|
|
|
# Custom Perl wrapper |
110
|
|
|
|
|
|
|
|
111
|
|
|
|
|
|
|
sub PDL::convolve{ |
112
|
1
|
|
|
1
|
0
|
7
|
my($x,$y,$c) = @_; |
113
|
1
|
50
|
33
|
|
|
13
|
barf("Usage: convolve(a(*), b(*), [o]c(*)") if $#_<1 || $#_>2; |
114
|
1
|
50
|
|
|
|
7
|
$c = PDL->null if $#_<2; |
115
|
1
|
50
|
|
|
|
6
|
&PDL::_convolve_int( $x->clump(-1), $y->clump(-1), |
116
|
|
|
|
|
|
|
long([$x->dims]), long([$y->dims]), |
117
|
|
|
|
|
|
|
($c->getndims>1? $c->clump(-1) : $c) |
118
|
|
|
|
|
|
|
); |
119
|
1
|
|
|
|
|
26
|
$c->setdims([$x->dims]); |
120
|
|
|
|
|
|
|
|
121
|
1
|
50
|
|
|
|
7
|
if($x->is_inplace) { |
122
|
0
|
|
|
|
|
0
|
$x .= $c; |
123
|
0
|
|
|
|
|
0
|
$x->set_inplace(0); |
124
|
0
|
|
|
|
|
0
|
return $x; |
125
|
|
|
|
|
|
|
} |
126
|
1
|
|
|
|
|
5
|
return $c; |
127
|
|
|
|
|
|
|
} |
128
|
|
|
|
|
|
|
|
129
|
|
|
|
|
|
|
|
130
|
|
|
|
|
|
|
|
131
|
|
|
|
|
|
|
*convolve = \&PDL::convolve; |
132
|
|
|
|
|
|
|
|
133
|
|
|
|
|
|
|
|
134
|
|
|
|
|
|
|
|
135
|
|
|
|
|
|
|
|
136
|
|
|
|
|
|
|
=head2 ninterpol() |
137
|
|
|
|
|
|
|
|
138
|
|
|
|
|
|
|
=for ref |
139
|
|
|
|
|
|
|
|
140
|
|
|
|
|
|
|
N-dimensional interpolation routine |
141
|
|
|
|
|
|
|
|
142
|
|
|
|
|
|
|
=for sig |
143
|
|
|
|
|
|
|
|
144
|
|
|
|
|
|
|
Signature: ninterpol(point(),data(n),[o]value()) |
145
|
|
|
|
|
|
|
|
146
|
|
|
|
|
|
|
=for usage |
147
|
|
|
|
|
|
|
|
148
|
|
|
|
|
|
|
$value = ninterpol($point, $data); |
149
|
|
|
|
|
|
|
|
150
|
|
|
|
|
|
|
C uses C to find a linearly interpolated value in |
151
|
|
|
|
|
|
|
N dimensions, assuming the data is spread on a uniform grid. To use |
152
|
|
|
|
|
|
|
an arbitrary grid distribution, need to find the grid-space point from |
153
|
|
|
|
|
|
|
the indexing scheme, then call C -- this is far from |
154
|
|
|
|
|
|
|
trivial (and ill-defined in general). |
155
|
|
|
|
|
|
|
|
156
|
|
|
|
|
|
|
See also L, which includes boundary |
157
|
|
|
|
|
|
|
conditions and allows you to switch the method of interpolation, but |
158
|
|
|
|
|
|
|
which runs somewhat slower. |
159
|
|
|
|
|
|
|
|
160
|
|
|
|
|
|
|
=cut |
161
|
|
|
|
|
|
|
|
162
|
|
|
|
|
|
|
|
163
|
|
|
|
|
|
|
*ninterpol = \&PDL::ninterpol; |
164
|
|
|
|
|
|
|
|
165
|
|
|
|
|
|
|
sub PDL::ninterpol { |
166
|
3
|
|
|
3
|
|
35
|
use PDL::Math 'floor'; |
|
3
|
|
|
|
|
9
|
|
|
3
|
|
|
|
|
38
|
|
167
|
3
|
|
|
3
|
|
24
|
use PDL::Primitive 'interpol'; |
|
3
|
|
|
|
|
17
|
|
|
3
|
|
|
|
|
36
|
|
168
|
0
|
0
|
|
0
|
0
|
0
|
print 'Usage: $x = ninterpolate($point(s), $data);' if $#_ != 1; |
169
|
0
|
|
|
|
|
0
|
my ($p, $y) = @_; |
170
|
0
|
|
|
|
|
0
|
my ($ip) = floor($p); |
171
|
|
|
|
|
|
|
# isolate relevant N-cube |
172
|
0
|
|
|
|
|
0
|
$y = $y->slice(join (',',map($_.':'.($_+1),list $ip))); |
173
|
0
|
|
|
|
|
0
|
for (list ($p-$ip)) { $y = interpol($_,$y->xvals,$y); } |
|
0
|
|
|
|
|
0
|
|
174
|
0
|
|
|
|
|
0
|
$y; |
175
|
|
|
|
|
|
|
} |
176
|
|
|
|
|
|
|
|
177
|
|
|
|
|
|
|
|
178
|
|
|
|
|
|
|
|
179
|
|
|
|
|
|
|
|
180
|
|
|
|
|
|
|
|
181
|
|
|
|
|
|
|
=head2 rebin |
182
|
|
|
|
|
|
|
|
183
|
|
|
|
|
|
|
=for sig |
184
|
|
|
|
|
|
|
|
185
|
|
|
|
|
|
|
Signature: (a(m); [o]b(n); int ns => n) |
186
|
|
|
|
|
|
|
|
187
|
|
|
|
|
|
|
=for ref |
188
|
|
|
|
|
|
|
|
189
|
|
|
|
|
|
|
N-dimensional rebinning algorithm |
190
|
|
|
|
|
|
|
|
191
|
|
|
|
|
|
|
=for usage |
192
|
|
|
|
|
|
|
|
193
|
|
|
|
|
|
|
$new = rebin $x, $dim1, $dim2,..;. |
194
|
|
|
|
|
|
|
$new = rebin $x, $template; |
195
|
|
|
|
|
|
|
$new = rebin $x, $template, {Norm => 1}; |
196
|
|
|
|
|
|
|
|
197
|
|
|
|
|
|
|
Rebin an N-dimensional array to newly specified dimensions. |
198
|
|
|
|
|
|
|
Specifying `Norm' keeps the sum constant, otherwise the intensities |
199
|
|
|
|
|
|
|
are kept constant. If more template dimensions are given than for the |
200
|
|
|
|
|
|
|
input pdl, these dimensions are created; if less, the final dimensions |
201
|
|
|
|
|
|
|
are maintained as they were. |
202
|
|
|
|
|
|
|
|
203
|
|
|
|
|
|
|
So if C<$x> is a 10 x 10 pdl, then C is a 15 x 10 pdl, |
204
|
|
|
|
|
|
|
while C is a 15 x 16 x 17 pdl (where the values |
205
|
|
|
|
|
|
|
along the final dimension are all identical). |
206
|
|
|
|
|
|
|
|
207
|
|
|
|
|
|
|
Expansion is performed by sampling; reduction is performed by averaging. |
208
|
|
|
|
|
|
|
If you want different behavior, use L |
209
|
|
|
|
|
|
|
instead. PDL::Transform::map runs slower but is more flexible. |
210
|
|
|
|
|
|
|
|
211
|
|
|
|
|
|
|
|
212
|
|
|
|
|
|
|
|
213
|
|
|
|
|
|
|
=for bad |
214
|
|
|
|
|
|
|
|
215
|
|
|
|
|
|
|
rebin does not process bad values. |
216
|
|
|
|
|
|
|
It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles. |
217
|
|
|
|
|
|
|
|
218
|
|
|
|
|
|
|
|
219
|
|
|
|
|
|
|
=cut |
220
|
|
|
|
|
|
|
|
221
|
|
|
|
|
|
|
|
222
|
|
|
|
|
|
|
|
223
|
|
|
|
|
|
|
|
224
|
|
|
|
|
|
|
|
225
|
|
|
|
|
|
|
|
226
|
|
|
|
|
|
|
# Custom Perl wrapper |
227
|
|
|
|
|
|
|
|
228
|
|
|
|
|
|
|
sub PDL::rebin { |
229
|
1
|
|
|
1
|
0
|
8
|
my($x) = shift; |
230
|
1
|
50
|
|
|
|
5
|
my($opts) = ref $_[-1] eq "HASH" ? pop : {}; |
231
|
1
|
|
|
|
|
5
|
my(@idims) = $x->dims; |
232
|
1
|
50
|
|
|
|
5
|
my(@odims) = ref $_[0] ? $_[0]->dims : @_; |
233
|
1
|
|
|
|
|
4
|
my($i,$y); |
234
|
1
|
|
|
|
|
3
|
foreach $i (0..$#odims) { |
235
|
2
|
50
|
|
|
|
10
|
if ($i > $#idims) { # Just dummy extra dimensions |
|
|
50
|
|
|
|
|
|
236
|
0
|
|
|
|
|
0
|
$x = $x->dummy($i,$odims[$i]); |
237
|
0
|
|
|
|
|
0
|
next; |
238
|
|
|
|
|
|
|
# rebin_int can cope with all cases, but code |
239
|
|
|
|
|
|
|
# 1->n and n->1 separately for speed |
240
|
|
|
|
|
|
|
} elsif ($odims[$i] != $idims[$i]) { # If something changes |
241
|
2
|
50
|
|
|
|
8
|
if (!($odims[$i] % $idims[$i])) { # Cells map 1 -> n |
|
|
50
|
|
|
|
|
|
242
|
0
|
|
|
|
|
0
|
my ($r) = $odims[$i]/$idims[$i]; |
243
|
0
|
|
|
|
|
0
|
$y = $x->mv($i,0)->dummy(0,$r)->clump(2); |
244
|
|
|
|
|
|
|
} elsif (!($idims[$i] % $odims[$i])) { # Cells map n -> 1 |
245
|
2
|
|
|
|
|
6
|
my ($r) = $idims[$i]/$odims[$i]; |
246
|
2
|
|
|
|
|
12
|
$x = $x->mv($i,0); |
247
|
|
|
|
|
|
|
# -> copy so won't corrupt input PDL |
248
|
2
|
|
|
|
|
35
|
$y = $x->slice("0:-1:$r")->copy; |
249
|
2
|
|
|
|
|
15
|
foreach (1..$r-1) { |
250
|
2
|
|
|
|
|
10
|
$y += $x->slice("$_:-1:$r"); |
251
|
|
|
|
|
|
|
} |
252
|
2
|
|
|
|
|
8
|
$y /= $r; |
253
|
|
|
|
|
|
|
} else { # Cells map n -> m |
254
|
0
|
|
|
|
|
0
|
&PDL::_rebin_int($x->mv($i,0), $y = null, $odims[$i]); |
255
|
|
|
|
|
|
|
} |
256
|
2
|
|
|
|
|
18
|
$x = $y->mv(0,$i); |
257
|
|
|
|
|
|
|
} |
258
|
|
|
|
|
|
|
} |
259
|
1
|
50
|
33
|
|
|
9
|
if (exists $opts->{Norm} and $opts->{Norm}) { |
260
|
1
|
|
|
|
|
4
|
my ($norm) = 1; |
261
|
1
|
|
|
|
|
3
|
for $i (0..$#odims) { |
262
|
2
|
50
|
|
|
|
5
|
if ($i > $#idims) { |
263
|
0
|
|
|
|
|
0
|
$norm /= $odims[$i]; |
264
|
|
|
|
|
|
|
} else { |
265
|
2
|
|
|
|
|
6
|
$norm *= $idims[$i]/$odims[$i]; |
266
|
|
|
|
|
|
|
} |
267
|
|
|
|
|
|
|
} |
268
|
1
|
|
|
|
|
104
|
return $x * $norm; |
269
|
|
|
|
|
|
|
} else { |
270
|
|
|
|
|
|
|
# Explicit copy so i) can't corrupt input PDL through this link |
271
|
|
|
|
|
|
|
# ii) don't waste space on invisible elements |
272
|
0
|
|
|
|
|
0
|
return $x -> copy; |
273
|
|
|
|
|
|
|
} |
274
|
|
|
|
|
|
|
} |
275
|
|
|
|
|
|
|
|
276
|
|
|
|
|
|
|
|
277
|
|
|
|
|
|
|
*rebin = \&PDL::rebin; |
278
|
|
|
|
|
|
|
|
279
|
|
|
|
|
|
|
|
280
|
|
|
|
|
|
|
|
281
|
|
|
|
|
|
|
|
282
|
|
|
|
|
|
|
=head2 circ_mean_p |
283
|
|
|
|
|
|
|
|
284
|
|
|
|
|
|
|
=for ref |
285
|
|
|
|
|
|
|
|
286
|
|
|
|
|
|
|
Calculates the circular mean of an n-dim image and returns |
287
|
|
|
|
|
|
|
the projection. Optionally takes the center to be used. |
288
|
|
|
|
|
|
|
|
289
|
|
|
|
|
|
|
=for usage |
290
|
|
|
|
|
|
|
|
291
|
|
|
|
|
|
|
$cmean=circ_mean_p($im); |
292
|
|
|
|
|
|
|
$cmean=circ_mean_p($im,{Center => [10,10]}); |
293
|
|
|
|
|
|
|
|
294
|
|
|
|
|
|
|
=cut |
295
|
|
|
|
|
|
|
|
296
|
|
|
|
|
|
|
|
297
|
|
|
|
|
|
|
sub circ_mean_p { |
298
|
0
|
|
|
0
|
1
|
0
|
my ($x,$opt) = @_; |
299
|
0
|
|
|
|
|
0
|
my ($rad,$sum,$norm); |
300
|
|
|
|
|
|
|
|
301
|
0
|
0
|
|
|
|
0
|
if (defined $opt) { |
302
|
0
|
|
|
|
|
0
|
$rad = long PDL::rvals($x,$opt); |
303
|
|
|
|
|
|
|
} |
304
|
|
|
|
|
|
|
else { |
305
|
0
|
|
|
|
|
0
|
$rad = long rvals $x; |
306
|
|
|
|
|
|
|
} |
307
|
0
|
|
|
|
|
0
|
$sum = zeroes($rad->max+1); |
308
|
0
|
|
|
|
|
0
|
PDL::indadd $x->clump(-1), $rad->clump(-1), $sum; # this does the real work |
309
|
0
|
|
|
|
|
0
|
$norm = zeroes($rad->max+1); |
310
|
0
|
|
|
|
|
0
|
PDL::indadd pdl(1), $rad->clump(-1), $norm; # equivalent to get norm |
311
|
0
|
|
|
|
|
0
|
$sum /= $norm; |
312
|
0
|
|
|
|
|
0
|
return $sum; |
313
|
|
|
|
|
|
|
} |
314
|
|
|
|
|
|
|
|
315
|
|
|
|
|
|
|
=head2 circ_mean |
316
|
|
|
|
|
|
|
|
317
|
|
|
|
|
|
|
=for ref |
318
|
|
|
|
|
|
|
|
319
|
|
|
|
|
|
|
Smooths an image by applying circular mean. |
320
|
|
|
|
|
|
|
Optionally takes the center to be used. |
321
|
|
|
|
|
|
|
|
322
|
|
|
|
|
|
|
=for usage |
323
|
|
|
|
|
|
|
|
324
|
|
|
|
|
|
|
circ_mean($im); |
325
|
|
|
|
|
|
|
circ_mean($im,{Center => [10,10]}); |
326
|
|
|
|
|
|
|
|
327
|
|
|
|
|
|
|
=cut |
328
|
|
|
|
|
|
|
|
329
|
|
|
|
|
|
|
|
330
|
|
|
|
|
|
|
sub circ_mean { |
331
|
0
|
|
|
0
|
1
|
0
|
my ($x,$opt) = @_; |
332
|
0
|
|
|
|
|
0
|
my ($rad,$sum,$norm,$a1); |
333
|
|
|
|
|
|
|
|
334
|
0
|
0
|
|
|
|
0
|
if (defined $opt) { |
335
|
0
|
|
|
|
|
0
|
$rad = long PDL::rvals($x,$opt); |
336
|
|
|
|
|
|
|
} |
337
|
|
|
|
|
|
|
else { |
338
|
0
|
|
|
|
|
0
|
$rad = long rvals $x; |
339
|
|
|
|
|
|
|
} |
340
|
0
|
|
|
|
|
0
|
$sum = zeroes($rad->max+1); |
341
|
0
|
|
|
|
|
0
|
PDL::indadd $x->clump(-1), $rad->clump(-1), $sum; # this does the real work |
342
|
0
|
|
|
|
|
0
|
$norm = zeroes($rad->max+1); |
343
|
0
|
|
|
|
|
0
|
PDL::indadd pdl(1), $rad->clump(-1), $norm; # equivalent to get norm |
344
|
0
|
|
|
|
|
0
|
$sum /= $norm; |
345
|
0
|
|
|
|
|
0
|
$a1 = $x->clump(-1); |
346
|
0
|
|
|
|
|
0
|
$a1 .= $sum->index($rad->clump(-1)); |
347
|
|
|
|
|
|
|
|
348
|
0
|
|
|
|
|
0
|
return $x; |
349
|
|
|
|
|
|
|
} |
350
|
|
|
|
|
|
|
|
351
|
|
|
|
|
|
|
|
352
|
|
|
|
|
|
|
|
353
|
|
|
|
|
|
|
|
354
|
|
|
|
|
|
|
=head2 kernctr |
355
|
|
|
|
|
|
|
|
356
|
|
|
|
|
|
|
=for ref |
357
|
|
|
|
|
|
|
|
358
|
|
|
|
|
|
|
`centre' a kernel (auxiliary routine to fftconvolve) |
359
|
|
|
|
|
|
|
|
360
|
|
|
|
|
|
|
=for usage |
361
|
|
|
|
|
|
|
|
362
|
|
|
|
|
|
|
$kernel = kernctr($image,$smallk); |
363
|
|
|
|
|
|
|
fftconvolve($image,$kernel); |
364
|
|
|
|
|
|
|
|
365
|
|
|
|
|
|
|
kernctr centres a small kernel to emulate the behaviour of the direct |
366
|
|
|
|
|
|
|
convolution routines. |
367
|
|
|
|
|
|
|
|
368
|
|
|
|
|
|
|
=cut |
369
|
|
|
|
|
|
|
|
370
|
|
|
|
|
|
|
|
371
|
|
|
|
|
|
|
*kernctr = \&PDL::kernctr; |
372
|
|
|
|
|
|
|
|
373
|
|
|
|
|
|
|
sub PDL::kernctr { |
374
|
|
|
|
|
|
|
# `centre' the kernel, to match kernel & image sizes and |
375
|
|
|
|
|
|
|
# emulate convolve/conv2d. FIX: implement with phase shifts |
376
|
|
|
|
|
|
|
# in fftconvolve, with option tag |
377
|
2
|
50
|
|
2
|
0
|
41
|
barf "Must have image & kernel for kernctr" if $#_ != 1; |
378
|
2
|
|
|
|
|
16
|
my ($imag, $kern) = @_; |
379
|
2
|
|
|
|
|
24
|
my (@ni) = $imag->dims; |
380
|
2
|
|
|
|
|
11
|
my (@nk) = $kern->dims; |
381
|
2
|
50
|
|
|
|
12
|
barf "Kernel and image must have same number of dims" if $#ni != $#nk; |
382
|
2
|
|
|
|
|
16
|
my ($newk) = zeroes(double,@ni); |
383
|
2
|
|
|
|
|
11
|
my ($k,$n,$y,$d,$i,@stri,@strk,@b); |
384
|
2
|
|
|
|
|
13
|
for ($i=0; $i <= $#ni; $i++) { |
385
|
4
|
|
|
|
|
9
|
$k = $nk[$i]; |
386
|
4
|
|
|
|
|
8
|
$n = $ni[$i]; |
387
|
4
|
50
|
|
|
|
11
|
barf "Kernel must be smaller than image in all dims" if ($n < $k); |
388
|
4
|
|
|
|
|
17
|
$d = int(($k-1)/2); |
389
|
4
|
|
|
|
|
20
|
$stri[$i][0] = "0:$d,"; |
390
|
4
|
|
|
|
|
14
|
$strk[$i][0] = (-$d-1).":-1,"; |
391
|
4
|
50
|
|
|
|
20
|
$stri[$i][1] = $d == 0 ? '' : ($d-$k+1).':-1,'; |
392
|
4
|
50
|
|
|
|
21
|
$strk[$i][1] = $d == 0 ? '' : '0:'.($k-$d-2).','; |
393
|
|
|
|
|
|
|
} |
394
|
|
|
|
|
|
|
# kernel is split between the 2^n corners of the cube |
395
|
2
|
|
|
|
|
10
|
my ($nchunk) = 2 << $#ni; |
396
|
|
|
|
|
|
|
CHUNK: |
397
|
2
|
|
|
|
|
13
|
for ($i=0; $i < $nchunk; $i++) { |
398
|
8
|
|
|
|
|
18
|
my ($stri,$strk); |
399
|
8
|
|
|
|
|
23
|
for ($n=0, $y=$i; $n <= $#ni; $n++, $y >>= 1) { |
400
|
16
|
50
|
|
|
|
38
|
next CHUNK if $stri[$n][$y & 1] eq ''; |
401
|
16
|
|
|
|
|
32
|
$stri .= $stri[$n][$y & 1]; |
402
|
16
|
|
|
|
|
36
|
$strk .= $strk[$n][$y & 1]; |
403
|
|
|
|
|
|
|
} |
404
|
8
|
|
|
|
|
14
|
chop ($stri); chop ($strk); |
|
8
|
|
|
|
|
14
|
|
405
|
8
|
|
|
|
|
33
|
($t = $newk->slice($stri)) .= $kern->slice($strk); |
406
|
|
|
|
|
|
|
} |
407
|
2
|
|
|
|
|
23
|
$newk; |
408
|
|
|
|
|
|
|
} |
409
|
|
|
|
|
|
|
|
410
|
|
|
|
|
|
|
|
411
|
|
|
|
|
|
|
|
412
|
|
|
|
|
|
|
|
413
|
|
|
|
|
|
|
|
414
|
|
|
|
|
|
|
=head2 convolveND |
415
|
|
|
|
|
|
|
|
416
|
|
|
|
|
|
|
=for sig |
417
|
|
|
|
|
|
|
|
418
|
|
|
|
|
|
|
Signature: (k0(); SV *k; SV *aa; SV *a) |
419
|
|
|
|
|
|
|
|
420
|
|
|
|
|
|
|
|
421
|
|
|
|
|
|
|
=for ref |
422
|
|
|
|
|
|
|
|
423
|
|
|
|
|
|
|
Speed-optimized convolution with selectable boundary conditions |
424
|
|
|
|
|
|
|
|
425
|
|
|
|
|
|
|
=for usage |
426
|
|
|
|
|
|
|
|
427
|
|
|
|
|
|
|
$new = convolveND($x, $kernel, [ {options} ]); |
428
|
|
|
|
|
|
|
|
429
|
|
|
|
|
|
|
Conolve an array with a kernel, both of which are N-dimensional. |
430
|
|
|
|
|
|
|
|
431
|
|
|
|
|
|
|
If the kernel has fewer dimensions than the array, then the extra array |
432
|
|
|
|
|
|
|
dimensions are threaded over. There are options that control the boundary |
433
|
|
|
|
|
|
|
conditions and method used. |
434
|
|
|
|
|
|
|
|
435
|
|
|
|
|
|
|
The kernel's origin is taken to be at the kernel's center. If your |
436
|
|
|
|
|
|
|
kernel has a dimension of even order then the origin's coordinates get |
437
|
|
|
|
|
|
|
rounded up to the next higher pixel (e.g. (1,2) for a 3x4 kernel). |
438
|
|
|
|
|
|
|
This mimics the behavior of the earlier L and |
439
|
|
|
|
|
|
|
L routines, so convolveND is a drop-in |
440
|
|
|
|
|
|
|
replacement for them. |
441
|
|
|
|
|
|
|
|
442
|
|
|
|
|
|
|
|
443
|
|
|
|
|
|
|
The kernel may be any size compared to the image, in any dimension. |
444
|
|
|
|
|
|
|
|
445
|
|
|
|
|
|
|
The kernel and the array are not quite interchangeable (as in mathematical |
446
|
|
|
|
|
|
|
convolution): the code is inplace-aware only for the array itself, and |
447
|
|
|
|
|
|
|
the only allowed boundary condition on the kernel is truncation. |
448
|
|
|
|
|
|
|
|
449
|
|
|
|
|
|
|
convolveND is inplace-aware: say C to modify |
450
|
|
|
|
|
|
|
a variable in-place. You don't reduce the working memory that way -- only |
451
|
|
|
|
|
|
|
the final memory. |
452
|
|
|
|
|
|
|
|
453
|
|
|
|
|
|
|
OPTIONS |
454
|
|
|
|
|
|
|
|
455
|
|
|
|
|
|
|
Options are parsed by PDL::Options, so unique abbreviations are accepted. |
456
|
|
|
|
|
|
|
|
457
|
|
|
|
|
|
|
=over 3 |
458
|
|
|
|
|
|
|
|
459
|
|
|
|
|
|
|
=item boundary (default: 'truncate') |
460
|
|
|
|
|
|
|
|
461
|
|
|
|
|
|
|
The boundary condition on the array, which affects any pixel closer |
462
|
|
|
|
|
|
|
to the edge than the half-width of the kernel. |
463
|
|
|
|
|
|
|
|
464
|
|
|
|
|
|
|
The boundary conditions are the same as those accepted by |
465
|
|
|
|
|
|
|
L, because this option is passed directly |
466
|
|
|
|
|
|
|
into L. Useful options are 'truncate' (the |
467
|
|
|
|
|
|
|
default), 'extend', and 'periodic'. You can select different boundary |
468
|
|
|
|
|
|
|
conditions for different axes -- see L for more |
469
|
|
|
|
|
|
|
detail. |
470
|
|
|
|
|
|
|
|
471
|
|
|
|
|
|
|
The (default) truncate option marks all the near-boundary pixels as BAD if |
472
|
|
|
|
|
|
|
you have bad values compiled into your PDL and the array's badflag is set. |
473
|
|
|
|
|
|
|
|
474
|
|
|
|
|
|
|
=item method (default: 'auto') |
475
|
|
|
|
|
|
|
|
476
|
|
|
|
|
|
|
The method to use for the convolution. Acceptable alternatives are |
477
|
|
|
|
|
|
|
'direct', 'fft', or 'auto'. The direct method is an explicit |
478
|
|
|
|
|
|
|
copy-and-multiply operation; the fft method takes the Fourier |
479
|
|
|
|
|
|
|
transform of the input and output kernels. The two methods give the |
480
|
|
|
|
|
|
|
same answer to within double-precision numerical roundoff. The fft |
481
|
|
|
|
|
|
|
method is much faster for large kernels; the direct method is faster |
482
|
|
|
|
|
|
|
for tiny kernels. The tradeoff occurs when the array has about 400x |
483
|
|
|
|
|
|
|
more pixels than the kernel. |
484
|
|
|
|
|
|
|
|
485
|
|
|
|
|
|
|
The default method is 'auto', which chooses direct or fft convolution |
486
|
|
|
|
|
|
|
based on the size of the input arrays. |
487
|
|
|
|
|
|
|
|
488
|
|
|
|
|
|
|
=back |
489
|
|
|
|
|
|
|
|
490
|
|
|
|
|
|
|
NOTES |
491
|
|
|
|
|
|
|
|
492
|
|
|
|
|
|
|
At the moment there's no way to thread over kernels. That could/should |
493
|
|
|
|
|
|
|
be fixed. |
494
|
|
|
|
|
|
|
|
495
|
|
|
|
|
|
|
The threading over input is cheesy and should probably be fixed: |
496
|
|
|
|
|
|
|
currently the kernel just gets dummy dimensions added to it to match |
497
|
|
|
|
|
|
|
the input dims. That does the right thing tersely but probably runs slower |
498
|
|
|
|
|
|
|
than a dedicated threadloop. |
499
|
|
|
|
|
|
|
|
500
|
|
|
|
|
|
|
The direct copying code uses PP primarily for the generic typing: it includes |
501
|
|
|
|
|
|
|
its own threadloops. |
502
|
|
|
|
|
|
|
|
503
|
|
|
|
|
|
|
|
504
|
|
|
|
|
|
|
|
505
|
|
|
|
|
|
|
=for bad |
506
|
|
|
|
|
|
|
|
507
|
|
|
|
|
|
|
convolveND does not process bad values. |
508
|
|
|
|
|
|
|
It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles. |
509
|
|
|
|
|
|
|
|
510
|
|
|
|
|
|
|
|
511
|
|
|
|
|
|
|
=cut |
512
|
|
|
|
|
|
|
|
513
|
|
|
|
|
|
|
|
514
|
|
|
|
|
|
|
|
515
|
|
|
|
|
|
|
|
516
|
|
|
|
|
|
|
|
517
|
3
|
|
|
3
|
|
31
|
use PDL::Options; |
|
3
|
|
|
|
|
282
|
|
|
3
|
|
|
|
|
2090
|
|
518
|
|
|
|
|
|
|
|
519
|
|
|
|
|
|
|
# Perl wrapper conditions the data to make life easier for the PP sub. |
520
|
|
|
|
|
|
|
|
521
|
|
|
|
|
|
|
sub PDL::convolveND { |
522
|
6
|
|
|
6
|
0
|
32
|
my($a0,$k,$opt0) = @_; |
523
|
6
|
|
|
|
|
22
|
my $inplace = $a0->is_inplace; |
524
|
6
|
|
|
|
|
19
|
my $x = $a0->new_or_inplace; |
525
|
|
|
|
|
|
|
|
526
|
|
|
|
|
|
|
|
527
|
6
|
50
|
|
|
|
29
|
barf("convolveND: kernel (".join("x",$k->dims).") has more dims than source (".join("x",$x->dims).")\n") |
528
|
|
|
|
|
|
|
if($x->ndims < $k->ndims); |
529
|
|
|
|
|
|
|
|
530
|
|
|
|
|
|
|
|
531
|
|
|
|
|
|
|
# Coerce stuff all into the same type. Try to make sense. |
532
|
|
|
|
|
|
|
# The trivial conversion leaves dataflow intact (nontrivial conversions |
533
|
|
|
|
|
|
|
# don't), so the inplace code is OK. Non-inplace code: let the existing |
534
|
|
|
|
|
|
|
# PDL code choose what type is best. |
535
|
6
|
|
|
|
|
18
|
my $type; |
536
|
6
|
50
|
|
|
|
16
|
if($inplace) { |
537
|
0
|
|
|
|
|
0
|
$type = $a0->get_datatype; |
538
|
|
|
|
|
|
|
} else { |
539
|
6
|
|
|
|
|
20
|
my $z = $x->flat->index(0) + $k->flat->index(0); |
540
|
6
|
|
|
|
|
141
|
$type = $z->get_datatype; |
541
|
|
|
|
|
|
|
} |
542
|
6
|
|
|
|
|
25
|
$x = $x->convert($type); |
543
|
6
|
|
|
|
|
12
|
$k = $k->convert($type); |
544
|
|
|
|
|
|
|
|
545
|
|
|
|
|
|
|
|
546
|
|
|
|
|
|
|
## Handle options -- $def is a static variable so it only gets set up once. |
547
|
6
|
|
|
|
|
11
|
our $def; |
548
|
6
|
100
|
|
|
|
13
|
unless(defined($def)) { |
549
|
1
|
|
|
|
|
10
|
$def = new PDL::Options( { |
550
|
|
|
|
|
|
|
Method=>'a', |
551
|
|
|
|
|
|
|
Boundary=>'t' |
552
|
|
|
|
|
|
|
} |
553
|
|
|
|
|
|
|
); |
554
|
1
|
|
|
|
|
5
|
$def->minmatch(1); |
555
|
1
|
|
|
|
|
4
|
$def->casesens(0); |
556
|
|
|
|
|
|
|
} |
557
|
|
|
|
|
|
|
|
558
|
6
|
|
|
|
|
18
|
my $opt = $def->options(PDL::Options::ifhref($opt0)); |
559
|
|
|
|
|
|
|
|
560
|
|
|
|
|
|
|
### |
561
|
|
|
|
|
|
|
# If the kernel has too few dimensions, we thread over the other |
562
|
|
|
|
|
|
|
# dims -- this is the same as supplying the kernel with dummy dims of |
563
|
|
|
|
|
|
|
# order 1, so, er, we do that. |
564
|
6
|
50
|
|
|
|
42
|
$k = $k->dummy($x->dims - 1, 1) |
565
|
|
|
|
|
|
|
if($x->ndims > $k->ndims); |
566
|
6
|
|
|
|
|
19
|
my $kdims = pdl($k->dims); |
567
|
|
|
|
|
|
|
|
568
|
|
|
|
|
|
|
### |
569
|
|
|
|
|
|
|
# Decide whether to FFT or directly convolve: if we're in auto mode, |
570
|
|
|
|
|
|
|
# choose based on the relative size of the image and kernel arrays. |
571
|
|
|
|
|
|
|
my $fft = ( ($opt->{Method} =~ m/^a/i) ? |
572
|
|
|
|
|
|
|
( $x->nelem > 2500 and ($x->nelem) <= ($k->nelem * 500) ) : |
573
|
6
|
50
|
0
|
|
|
33
|
( $opt->{Method} !~ m/^[ds]/i ) |
574
|
|
|
|
|
|
|
); |
575
|
|
|
|
|
|
|
|
576
|
|
|
|
|
|
|
### |
577
|
|
|
|
|
|
|
# Pad the array to include boundary conditions |
578
|
6
|
|
|
|
|
15
|
my $adims = pdl($x->dims); |
579
|
6
|
|
|
|
|
347
|
my $koff = ($kdims/2)->ceil - 1; |
580
|
|
|
|
|
|
|
|
581
|
|
|
|
|
|
|
my $aa = $x->range( -$koff, $adims + $kdims, $opt->{Boundary} ) |
582
|
6
|
|
|
|
|
194
|
->sever; |
583
|
|
|
|
|
|
|
|
584
|
6
|
100
|
|
|
|
53
|
if($fft) { |
585
|
|
|
|
|
|
|
# The eval here keeps conflicts from happening at compile time |
586
|
3
|
|
|
1
|
|
248
|
eval "use PDL::FFT" ; |
|
1
|
|
|
1
|
|
541
|
|
|
1
|
|
|
1
|
|
4
|
|
|
1
|
|
|
|
|
7
|
|
|
1
|
|
|
|
|
8
|
|
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
4
|
|
|
1
|
|
|
|
|
9
|
|
|
1
|
|
|
|
|
3
|
|
|
1
|
|
|
|
|
5
|
|
587
|
|
|
|
|
|
|
|
588
|
3
|
50
|
|
|
|
17
|
print "convolveND: using FFT method\n" if($PDL::debug); |
589
|
|
|
|
|
|
|
|
590
|
|
|
|
|
|
|
# FFT works best on doubles; do our work there then cast back |
591
|
|
|
|
|
|
|
# at the end. |
592
|
3
|
|
|
|
|
11
|
$aa = double($aa); |
593
|
3
|
|
|
|
|
10
|
my $aai = $aa->zeroes; |
594
|
|
|
|
|
|
|
|
595
|
3
|
|
|
|
|
9
|
my $kk = $aa->zeroes; |
596
|
3
|
|
|
|
|
9
|
my $kki = $aa->zeroes; |
597
|
3
|
|
|
|
|
4
|
my $tmp; # work around new perl -d "feature" |
598
|
3
|
|
|
|
|
138
|
($tmp = $kk->range( - ($kdims/2)->floor, $kdims, 'p')) .= $k; |
599
|
3
|
|
|
|
|
39
|
PDL::fftnd($kk, $kki); |
600
|
3
|
|
|
|
|
10
|
PDL::fftnd($aa, $aai); |
601
|
|
|
|
|
|
|
|
602
|
|
|
|
|
|
|
{ |
603
|
3
|
|
|
|
|
4
|
my($ii) = $kk * $aai + $aa * $kki; |
|
3
|
|
|
|
|
141
|
|
604
|
3
|
|
|
|
|
102
|
$aa = $aa * $kk - $kki * $aai; |
605
|
3
|
|
|
|
|
17
|
$aai .= $ii; |
606
|
|
|
|
|
|
|
} |
607
|
|
|
|
|
|
|
|
608
|
3
|
|
|
|
|
12
|
PDL::ifftnd($aa,$aai); |
609
|
3
|
|
|
|
|
10
|
$x .= $aa->range( $koff, $adims); |
610
|
|
|
|
|
|
|
|
611
|
|
|
|
|
|
|
} else { |
612
|
3
|
50
|
|
|
|
9
|
print "convolveND: using direct method\n" if($PDL::debug); |
613
|
|
|
|
|
|
|
|
614
|
|
|
|
|
|
|
### The first argument is a dummy to set $GENERIC. |
615
|
3
|
|
|
|
|
10
|
&PDL::_convolveND_int( $k->flat->index(0), $k, $aa, $x ); |
616
|
|
|
|
|
|
|
|
617
|
|
|
|
|
|
|
} |
618
|
|
|
|
|
|
|
|
619
|
|
|
|
|
|
|
|
620
|
6
|
|
|
|
|
81
|
$x; |
621
|
|
|
|
|
|
|
} |
622
|
|
|
|
|
|
|
|
623
|
|
|
|
|
|
|
|
624
|
|
|
|
|
|
|
|
625
|
|
|
|
|
|
|
*convolveND = \&PDL::convolveND; |
626
|
|
|
|
|
|
|
|
627
|
|
|
|
|
|
|
|
628
|
|
|
|
|
|
|
|
629
|
|
|
|
|
|
|
; |
630
|
|
|
|
|
|
|
|
631
|
|
|
|
|
|
|
|
632
|
|
|
|
|
|
|
=head1 AUTHORS |
633
|
|
|
|
|
|
|
|
634
|
|
|
|
|
|
|
Copyright (C) Karl Glazebrook and Craig DeForest, 1997, 2003 |
635
|
|
|
|
|
|
|
All rights reserved. There is no warranty. You are allowed |
636
|
|
|
|
|
|
|
to redistribute this software / documentation under certain |
637
|
|
|
|
|
|
|
conditions. For details, see the file COPYING in the PDL |
638
|
|
|
|
|
|
|
distribution. If this file is separated from the PDL distribution, |
639
|
|
|
|
|
|
|
the copyright notice should be included in the file. |
640
|
|
|
|
|
|
|
|
641
|
|
|
|
|
|
|
=cut |
642
|
|
|
|
|
|
|
|
643
|
|
|
|
|
|
|
|
644
|
|
|
|
|
|
|
|
645
|
|
|
|
|
|
|
|
646
|
|
|
|
|
|
|
|
647
|
|
|
|
|
|
|
|
648
|
|
|
|
|
|
|
# Exit with OK status |
649
|
|
|
|
|
|
|
|
650
|
|
|
|
|
|
|
1; |
651
|
|
|
|
|
|
|
|
652
|
|
|
|
|
|
|
|