[Insight-users] Weird result from Point Set Registration

[junyixia]

Hi Luis,
    Thank you for your reply. Is there any guideline of choosing the 
optimization method for the point based registration? Thanks.

- Junyi

Luis Ibanez wrote:
>
>
> Hi Junyi,
>
>
> You *must* initialize the transform to a value closer
> to the real misalignment between the two point sets.
>
> The geometrical arrangement of points that you have
> is particularly prone to mis-correspondance between
> the points.
>
>
> As you probably know, pointset-to-pointset registration
> is actually trying two solve to separate but related
> problems:
>
>           1) Point correspondance
>           2) Registration
>
>
> If you don't initialize the Transform to a value closer
> to the real mis-alignment, then the point will probably
> find wrong correspondances that satisfy a local minimum
> in the sum of squared distances.
>
>
>
>    Regards,
>
>
>       Luis