Dear FSL users,
I'm wondering what exactly the cost weighting option in FLIRT does internally. My guess is that for registration, FSL has some kind of measure/error for every voxel which tells how good/bad that voxel is registered (this measure then depends on what cost function is chosen) and that by adding all these errors for all voxels you get a measure for the goodness of the total registration. This would mean that with voxel weights, you could influence this final measure, for instance by imposing high weights in areas that are important in the registration so that you can kind of force FSL to have lower errors in those areas. However, if it works that way, I don't get why there is the option to define both cost function weightings for the input as well as the reference image. At least, defining both wouldn't make sense would it? Could somebody explain this to me in addition to the functioning of cost function weighting?
Thanks in advance,
Joep Kamps
|