Data normalization by volume

Currently, when output data is cast to uint16 (ex. jpeg2000 file format), the normalization is done slice by slice.

It would be more accurate to compute the min/max on the whole volume (and ideally compute a histogram).