Hi again:
Ignore that last post. I missed the line where
my($a) = $in->(0:$d)->copy - $opt->{post};
...
$out->(0:$d) .= $a x $opt->{inverse} - $opt->{pre};
I'm still working on it...
--greg
Greg Ederer wrote:
Hi All:
The flags seem to be a non-issue. I am looking inside of
PDL::Transform.pm and am wondering about some things:
First, since I'm relatively new to matrix manipulation with PDL, I
expanded the "map" function to match Craig's description of what it is
supposed to do, and removed all the options so I wouldn't get
confused. Here's the transform part:
##############################
## Rubber meets the road: calculate the inverse transformed points.
my $ndc = PDL::Basic::ndcoords(@dd);
my $idx = $ndc->double + 0.5; # shift from
pixel corner to pixel center
$idx = PDL::Transform::t_fits($out)->apply($idx); # output
[col,row] => output scientific
$idx = $me->invert($idx); #
output scientific => input scientific
$idx = PDL::Transform::t_fits($in)->invert($idx); # input
scientific => input [col,row]
$a = $in->interpND($idx,{method=>'s', bound=>'t'}); #
interpolate values at [col,row]
$out->slice(':') .= $a; # trivial slice prevents header overwrite...
return $out;
########################
What I find is that the results of each transform stage look OK until
I get to the line
$idx = PDL::Transform::t_fits($in)->invert($idx); # input
scientific to input [col,row]
The resulting idx matrix has [col,row] that are not scaled correctly.
So I went and looked at the "t_fits" definition, which I see is a
subclass of "t_linear", which defines the following functions :
$me->{func} = sub {
.....
$out->(0:$d) .= $a x $opt->{matrix} + $opt->{post};
return $out;
};
and
$me->{inv} = sub {
....
$out->(0:$d) .= $a x $opt->{inverse} - $opt->{pre};
$out;
}
and my question is, shouldn't the inverse be defined as
$out->(0:$d) .= ($a - $opt->{pre}) x $opt->{inverse};
Again, my matrix math is rusty, but for a line,
y = mx + b;
so
x=(y-b) / m;
I haven't had time to try this yet, but please correct me if I'm
barking up the wrong tree.
Thanks,
--greg
_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
