Bug #64636 for Geometry-AffineTransform: untransform coordinates

Sun Jan 09 16:34:07 2011 user42 [...] zip.com.au - Ticket created

Subject:	untransform coordinates
Date:	Mon, 10 Jan 2011 08:34:05 +1100
To:	bug-Geometry-AffineTransform [...] rt.cpan.org
From:	Kevin Ryde <user42 [...] zip.com.au>

After transforming some coordinate values to screen pixel positions I thought to do the reverse, taking a pixel position back to original values. Is there an untransform hiding in the code?, like ($x,$y) = $aff->untransform($x,y) I suppose an ->invert() or something could invert the whole transform object. So maybe $rev=$aff->clone->invert to get a reverse. Though I think I only want a handful of untransforms. Incidentally, as a suggestion for the docs, I like seeing a sample call for methods rather than a description in words (or a sample as well as words). I find a sample easier to read quickly, cut and paste, etc.

Mon Jan 10 17:07:27 2011 user42 [...] zip.com.au - Correspondence added

Subject:	Re: [rt.cpan.org #64636] untransform coordinates
Date:	Tue, 11 Jan 2011 09:07:27 +1100
To:	bug-Geometry-AffineTransform [...] rt.cpan.org
From:	Kevin Ryde <user42 [...] zip.com.au>

Marc Liyanage <marc@entropy.ch> writes: Show quoted text

> > both are good suggestions, unfortunately I don't have time to work on > it anytime soon.

I might have a go at what I think I want. Show quoted text

> I looked into the inverse a while ago, and I wasn't sure how to do it > properly so all cases are handled.

It might be as easy as the matrix inverse (a division by the determinant, and taking the values swapped, or some such) and applying the offset either before or after, whichever is the right way around :) Show quoted text

> You're the first person ever other than me who I know using it. What > are you doing with it and how did you find the module?

I was experimenting in my math-image program (on cpan!) with a bit of a rotate, scale and offset. I'd had my own few lines for scale and offset, and had seen Transform::Canvas doing similar, then twigged to search for "affine" for rotate as well. I'm not sure how much rotating I'll end up wanting, but even for just scale and offset I was thinking "why can't someone else do it" :-)

Fri Jan 14 16:55:46 2011 user42 [...] zip.com.au - Correspondence added

Subject:	Re: [rt.cpan.org #64636] untransform coordinates
Date:	Sat, 15 Jan 2011 08:54:12 +1100
To:	bug-Geometry-AffineTransform [...] rt.cpan.org
From:	Kevin Ryde <user42 [...] zip.com.au>

I got to the few lines below. Haven't made actual tests, but seems pretty close. I might make a croak for det==0 instead of letting it throw divide by zero. (That arising from a scale==0 factor, either X or Y or both, and there being no way to get back original positions in that case.) I see the java class has longer name inverseTransform, but I think I like untransform better. # ($x,$y, $x,$y, ...) = $aff->untransform($x,$y, $x,$y, ...) sub untransform { my $self = shift; my @result; my $det = $self->{m11}*$self->{m22} - $self->{m12}*$self->{m21}; while (@_) { my $x = shift() - $self->{tx}; my $y = shift() - $self->{ty}; push @result, ($self->{m22} * $x - $self->{m21} * $y) / $det, ($self->{m11} * $y - $self->{m12} * $x) / $det; } return @result; } # $aff->invert, and returning the modified $aff sub invert { my ($self) = @_; my $det = $self->{m11}*$self->{m22} - $self->{m12}*$self->{m21}; return $self->set_matrix_2x3 ($self->{m22} / $det, # 11 - $self->{m12} / $det, # 12 - $self->{m21} / $det, # 21 $self->{m11} / $det, # 22 $self->untransform(0,0)); # tx,ty as full expressions instead of untransform(), if preferred # ($self->{m21} * $self->{ty} - $self->{m22} * $self->{tx}) / $det, # ($self->{m12} * $self->{tx} - $self->{m11} * $self->{ty}) / $det); }

Wed Jan 19 14:10:25 2011 marc [...] entropy.ch - Correspondence added

Subject:	Re: [rt.cpan.org #64636] untransform coordinates
Date:	Wed, 19 Jan 2011 11:10:14 -0800
To:	bug-Geometry-AffineTransform [...] rt.cpan.org
From:	Marc Liyanage <marc [...] entropy.ch>

Thanks for this implementation, that's great. I added an invert() method to the class, everything else can be composed with that method: $t->invert()->transform(...) $t->clone()->invert()->transform(...) etc. I added the exception when the determinant is zero, and some new unit tests to test it all. Thanks again. On Jan 14, 2011, at 13:55 , Kevin Ryde via RT wrote: Show quoted text

> Queue: Geometry-AffineTransform > Ticket <URL: https://rt.cpan.org/Ticket/Display.html?id=64636 > > > I got to the few lines below. Haven't made actual tests, but seems > pretty close. > > I might make a croak for det==0 instead of letting it throw divide by > zero. (That arising from a scale==0 factor, either X or Y or both, and > there being no way to get back original positions in that case.) > > I see the java class has longer name inverseTransform, but I think I > like untransform better. > > > > # ($x,$y, $x,$y, ...) = $aff->untransform($x,$y, $x,$y, ...) > sub untransform { > my $self = shift; > my @result; > my $det = $self->{m11}*$self->{m22} - $self->{m12}*$self->{m21}; > while (@_) { > my $x = shift() - $self->{tx}; > my $y = shift() - $self->{ty}; > push @result, > ($self->{m22} * $x - $self->{m21} * $y) / $det, > ($self->{m11} * $y - $self->{m12} * $x) / $det; > } > return @result; > } > > # $aff->invert, and returning the modified $aff > sub invert { > my ($self) = @_; > my $det = $self->{m11}*$self->{m22} - $self->{m12}*$self->{m21}; > return $self->set_matrix_2x3 > ($self->{m22} / $det, # 11 > - $self->{m12} / $det, # 12 > - $self->{m21} / $det, # 21 > $self->{m11} / $det, # 22 > $self->untransform(0,0)); > > # tx,ty as full expressions instead of untransform(), if preferred > # ($self->{m21} * $self->{ty} - $self->{m22} * $self->{tx}) / $det, > # ($self->{m12} * $self->{tx} - $self->{m11} * $self->{ty}) / $det); > } >

Show quoted text

______________________________ Marc Liyanage www.entropy.ch skype mliyanage iChat liyanage@mac.com

Wed Jan 19 14:10:26 2011 The RT System itself - Status changed from 'new' to 'open'

Wed Jan 19 20:47:25 2011 reg.cpan [...] entropy.ch - Correspondence added

Fixed in 1.3

Wed Jan 19 20:47:25 2011 reg.cpan [...] entropy.ch - Status changed from 'open' to 'resolved'

Wed Jan 19 20:47:26 2011 reg.cpan [...] entropy.ch - Given to LIYANAGE

Tue Feb 01 19:55:12 2011 user42 [...] zip.com.au - Correspondence added

Subject:	Re: [rt.cpan.org #64636] untransform coordinates
Date:	Wed, 02 Feb 2011 11:55:26 +1100
To:	bug-Geometry-AffineTransform [...] rt.cpan.org
From:	Kevin Ryde <user42 [...] zip.com.au>

"Marc Liyanage via RT" <bug-Geometry-AffineTransform@rt.cpan.org> writes: Show quoted text

> > $t->clone()->invert()->transform(...)

And the untransform() method to do that too? You don't want the java class having more ways to do it than perl! :) The untransform() doesn't have to be particularly fast. I don't mind that it might calculate the determinant each time. You could leave the comment in invert() that the inverted tx,ty are the 0,0 point untransformed, as it's a bit hairy.

Tue Feb 01 19:55:13 2011 The RT System itself - Status changed from 'resolved' to 'open'