[Insight-users] Re: On incorrect matrix inverse

[Suyash P. Awate]

Hi everybody,

Thanks for the generous help.

I was also trying to get the inverse concerning with the registration application, like Fucang (see email attached).

Here is my piece of code that gave me the wrong result:

------------
const unsigned int Dimension = 3;
typedef double PixelType;
typedef itk::Matrix <PixelType, Dimension, Dimension> MatrixType;
MatrixType matrix;
  matrix.Fill (0);
  matrix (0,0)= atof (argv[4]);
  matrix (0,1)= atof (argv[5]);
  matrix (0,2)= atof (argv[6]);
  matrix (1,0)= atof (argv[7]);
  matrix (1,1)= atof (argv[8]);
  matrix (1,2)= atof (argv[9]);
  matrix (2,0)= atof (argv[10]);
  matrix (2,1)= atof (argv[11]);
  matrix (2,2)= atof (argv[12]);
vnl_matrix_fixed <PixelType, Dimension, Dimension> inverse;
inverse= matrix.GetInverse();
-----------
matrix
 0.99925   0.0371382 -0.0109982
-0.0373772 0.99905   -0.0223916
 0.0101562 0.0227859  0.999689

warnings:
InsightToolkit-2.2.0/Utilities/vxl/core/vnl/algo/vnl_svd.txx: suspicious return value (2) from SVDC
InsightToolkit-2.2.0/Utilities/vxl/core/vnl/algo/vnl_svd.txx: M is 3x3
M = [ ...
 0.9992500000000  0.0371382000000 -0.0109982000000
-0.0373772000000  0.9990500000000 -0.0223916000000
 0.0101562000000  0.0227859000000  0.9996890000000  ]

matrix inverse using vnl in itk (incorrect)
-0.982557 -0.0644525 -0.174434
-0.0677511 0.997618   0.0130162
-0.17318  -0.0246073  0.984583

matrix inverse in matlab (correct)
 0.9992   -0.0374    0.0102
 0.0371    0.9991    0.0228
-0.0110   -0.0224    0.9997
------------

I had also tried using the AffineTransform but for some reason could not get that to work and so I wrote this simple matrix code. Perhaps that was due to the same problem that Fucang reported.
Currently, however, my problem is different. As we can see I am not using any Transform class at all. Moreover I am getting warnings from the vnl_svd class, so vnl is definitely involved here.

Julien: I am giving 9 numbers (truncated: 0.99925   0.0371382 -0.0109982 -0.0373772 0.99905 -0.0223916 0.0101562 0.0227859  0.999689) to the matrix in itk and also the same numbers to the matrix in matlab. I got these numbers from the output of a separate program.
Just FYI, I have compiled ITK on Suse 10.0 using GCC 4.0. I wonder if that has anything to do with this !

Best regards,
Suyash.


  ----- Original Message ----- 
  From: jiafucang 
  To: suyash at cs.utah.edu ; jjomier at cs.unc.edu 
  Sent: Friday, November 11, 2005 7:10 AM
  Subject: On incorrect matrix inverse


  Hello, Suyash, Jomier,

  I have reported similar problem on
  http://public.kitware.com/pipermail/insight-users/2005-October/015441.html 

  and added a bug 2450

  http://www.itk.org/Bug/bug.php?op=show&bugid=2450&pos=24

  I think it is because ITK does not computer inverse when setparameters used, and not VNL problem.

  HTH

  Fucang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20051111/a581dd41/attachment.html