Re: [OctDev] imrotate

[Justus P. <Jus...@UL...>]

Paul Kienzle <pki...@us...> wrote on Tue, 22 Feb 2005
07:13:43 -0500:

> Justus,
>
> I was testing the wrong code.  When I look at it now it looks fine.
> Sorry to trouble you.

Frankly, it did trouble me ("imrotate seems to be broken for all
of...", "particularly glaring", "Can the [other!?] users of this
function please come up with a fix"). But I accept your apology.

> I would still like you to tell me which tests you prefer.  I know I've
> thrown a lot of suggestions your way :-)

I went through all of the suggestions and cast comprehensive versions
of the essence of them into the following three tests, which I suggest
to be added to the end of imrotate.m.

Justus


%!test
%! ## Verify minimal loss across six rotations that add up to 360 +/- 1 deg=
.:
%! methods =3D { "nearest", "bilinear", "bicubic", "Fourier" };
%! angles     =3D [ 59  60  61  ];
%! tolerances =3D [ 7.4 8.5 8.6	  # nearest
%!                3.5 3.1 3.5     # bilinear
%!                2.7 0.1 2.7     # bicubic
%!                2.7 1.6 2.8 ];  # Fourier
%! x =3D peaks(50);
%! x -=3D min(min(x));	      # Fourier does not handle neg. values well
%! for m =3D 1:(length(methods))
%!   y =3D x;
%!   for i =3D 1:5
%!     y =3D imrotate(y, 60, methods(m), "crop");
%!   end
%!   for a =3D 1:(length(angles))
%!     assert(norm((x - imrotate(y, angles(a), methods(m), "crop"))
%!                 (10:40, 10:40)) < tolerances(m,a));
%!   end
%! end


%!test
%! ## Verify exactness of near-90 and 90-degree rotations:
%! X =3D rand(99);
%! for angle =3D [90 180 270]
%!   for da =3D [-0.1 0.1]
%!     Y =3D imrotate(X,   angle + da , "nearest");
%!     Z =3D imrotate(Y, -(angle + da), "nearest");
%!     assert(norm(X - Z) =3D=3D 0); # exact zero-sum rotation
%!     assert(norm(Y - imrotate(X, angle, "nearest")) =3D=3D 0); # near zer=
o-sum
%!   end
%! end


%!test
%! ## Verify preserved pixel density:
%! methods =3D { "nearest", "bilinear", "bicubic", "Fourier" };
%! ## This test does not seem to do justice to the Fourier method...:
%! tolerances =3D [ 4 2.2 2.0 209 ];
%! range =3D 3:9:100;
%! for m =3D 1:(length(methods))
%!   t =3D [];
%!   for n =3D range
%!     t(end + 1) =3D sum(imrotate(eye(n), 20, methods(m))(:));
%!   end
%!   assert(t, range, tolerances(m));
%! end


--=20
Justus H. Piater, Ph.D.         http://www.montefiore.ulg.ac.be/~piater/
Institut Montefiore, B28        Phone: +32-4-366-2279
Universit=E9 de Li=E8ge, Belgium    Fax:   +32-4-366-2620