Returns a focal fn(shifts) computing the classic bilateral filter:
each output pixel is the window mean weighted by a spatial Gaussian
(distance from the centre, sigma_d) times a range Gaussian
(difference from the centre VALUE, sigma_r), so smoothing stays
within regions of similar value and stops at sharp transitions. Use
as focal(x, fn = bilateral_focal(sigma_r), radius = 1L).
Arguments
- sigma_r
Range Gaussian standard deviation (data units).
- sigma_d
Spatial Gaussian standard deviation in pixels (default 1, matching the default 3x3 window).
- radius
Window radius the body is built for; must match the
radiuspassed tofocal()(default 1 = 3x3).
Value
A focal body fn(shifts) for focal().
Details
A NaN centre stays NaN; NaN neighbours (and the NaN halo garry pads
outside the raster) drop out of the weighted mean (the semantics of
rustyfilters::rf_bilateral(edge = "shrink", na_policy = "omit")).
sigma_r must be supplied: the parameter-free per-band default
(the band's own sd) is a whole-raster statistic, so compute it in a
separate reduce pass (or reuse fitted values) and pass it in.