[Insight-users] slice to volume registration: optimization is correct?

[stan]

Hi all,

I'm doing some tests on slice to volume affine registration.
To test my code, I register a specific slice of the 
volume (fixed image) with the whole volume (moving image) by
providing an initial transformation (an affine transformation 
which only rotates the slice 10 deg around the x-axis).

Even when the result is fine, I've found that some parameters
are not optimized at all ... (see the 6th and the 9th parameters):

Initial parameters = [1, 0, 0, 0, 0.984808, 0.173649, 0, -0.173649,
0.984808, 0, 0, 0]

0   0.2   -0.557382   [1.00186, -0.000366774, 0, 0.00025731, 0.986009,
0.173649, -0.000127247, -0.167419, 0.984808, -0.0105518, 0.196942,
-0.032534]

1   0.2   -0.577705   [0.998905, 0.000928042, 0, 0.000672959, 0.986261,
0.173649, -0.000406448, -0.159618, 0.984808, -0.0102843, 0.391734,
-0.077084]

2   0.2   -0.595902   [1.0119, 0.000218071, 0, 0.00082287, 0.989483,
0.173649, -8.3977e-05, -0.15157, 0.984808, 0.00914714, 0.589726,
-0.0903846]

...

77   0.000195313   -1.64248   [0.999986, -2.18366e-05, 0, -2.51637e-06,
0.999994, 0.173649, -4.839e-05, -3.68753e-05, 0.984808, 0.00165371,
0.00172779, 0.00317447]

Final parameters = [0.999986, -2.18366e-05, 0, -2.51637e-06, 0.999994,
0.173649, -4.839e-05, -3.68753e-05, 0.984808, 0.00165371, 0.00172779,
0.00317447]

I wonder if this is an expected possible behaviour of the optimizer, and
if so, how to explain this ... I would have expected to obtain a result
equal to the identity, as I obtain when the initial rotation is around 
the z-axis ...

I've attached my code (I've removed the irrelevant parts of it). 
I run it with rx=-10*3.1416/180 and ry=rz=tx=ty=tz=0.


Thanks !

Stan




-------------- next part --------------
A non-text attachment was scrubbed...
Name: CodeForITK.cxx
Type: text/x-c++src
Size: 6306 bytes
Desc: not available
URL: <http://www.itk.org/pipermail/insight-users/attachments/20100108/da0ba0b4/attachment.cxx>