Registration: Fix local cross-correlation metric

[Lestropie]

No idea if this is even accessible from the command-line, but worth fixing nevertheless in case anyone out there is trying to do anything with it.

[Lestropie]

I discovered a second potential minor issue with the LCC code.

Here:

vector<uint32_t> NoOversample;

This empty class instance is named with the intention of resulting in no oversampling during reslicing. It is passed to the Adapter::Reslice<> constructor:

else if (parameters.im1_mask.valid() and parameters.im2_mask.valid()){
  Adapter::Reslice<Interp::Nearest, Im1MaskType> mask_reslicer1 (parameters.im1_mask, cc_mask_header, parameters.transformation.get_transform_half(), NoOversample);
  Adapter::Reslice<Interp::Nearest, Im2MaskType> mask_reslicer2 (parameters.im2_mask, cc_mask_header, parameters.transformation.get_transform_half_inverse(), NoOversample);

That class however defines:

extern const vector<uint32_t> AutoOverSample;

as an equivalent empty class instance. So I believe in the LCC code the masks are being automatically oversampled despite the variable naming indicating an intent to the contrary.

The subsequent code however looks like it's entirely compatible with oversampling of mask images resulting in continuous values:

auto both = [](decltype(cc_mask)& cc_mask, decltype(mask_reslicer1)& m1, decltype(mask_reslicer2)& m2) {
  cc_mask.value() = ((m1.value() + m2.value()) / 2.0) > 0.5 ? true : false;
};

So I think just remove that NoOversample variable and leave the behaviour as the default AutoOverSample?