To follow on @whuber, it would help to have a little more clarification on what's being eliminated.
Cleaning up prior to rasterization could be much more efficient. Not sure the best way with the raw point clouds, but creating a TIN might be at least lighter weight solution than a raster. One could check for slopes exceeding a threshold, deal with the associated points, and then rasterizing using the cleaned up point set.
As a pre-process to cleaning up after rasterizing, one could use @radouxju's suggestion to create masks of ridges which one could then assess for what you're really after. The
basin function (doc) might also be helpful. Probably want to use something like
Curvature (doc) or threshold against the slope of a slope surface (i.e., the 2nd order derivative of the surface) to find peaks that are internal to basins. With basins and to some extent with sinks, you can use focalvariety to find where zones abut, extract line-style boundaries to further limit the spatial extent of your subsequent analysis if that makes sense. I'm guessing that you're not really looking to find real geomorphic features (as @whuber suggests), so this may not be so helpful.
I would look at using focal statistics to try to identify problem cells or groups of cells (using a neighborhood). It's pretty versatile
There's a lot of stuff out there in the remote sensing regarding pass filters (like high and low frequency ones), but this doesn't really sound like what you're after.
Again, kind of guessing at what the OP is after.