[Insight-developers] Problems with Affine Registration

[Paul Hughett]

> What kind of results are you getting?  Is it doing nothing at all or
> is it converging to an unexpected solution?

I've been working most recently with the
ImageToImageAffineMutualInformationGradientDescentRegistrationTest
program, modified to add a scale factor within the reference gaussian,
e.g.

    ri.Set( 200.0 * exp( - 1.1*( x*x + y*y )/(s*s) ) );
                           ^^^^
This corresponds to roughly a 5% isotropic scaling, and should yield
coefficients of [0.95, 0.95] (or maybe [1.05, 1.05]) on the diagonal
of the affine transform matrix.  What actually happens is that the
diagonal stays within 0.001 or 0.0001 of unity.

> What happens when you fiddle with the number of iterations and/or the
> learning rates?

Doesn't seem to make any difference within the range of values that
I've tried: roughly up to 1000 on the number of iterations, and from
0.0001 to 0.1 on the learning rate.  The larger learning rates seem to
cause divergence (the images do not overlap any more!).


Paul Hughett