How can I repopulate raster regions with value of larger adjacent region

[thomaswarnaar]

Dear all,

I need to repopulate regions under a specific total area threshold with the value of larger adjacent regions (in order to generalize a map). Up to now I've used ESRI's suggested workflow for classified aerial imagery generalization, but I would like to try to keep existing boundaries intact. In the ESRI workflow in the end the Nibble tool is used using a mask where I extracted regions larger than the minimum area, however Nibble destroys the original boundaries. 

So I would like to fill in the smaller (unmasked so to say) regions with the values of larger adjacent regions. But not by repopulating the unselected (smaller) regions completely with one adjacent value as within the extraction original boundaries may be neglected. For example if there are two or more regions next to each other all smaller than the minimum mapping unit I want them filled while keeping their original boundaries in mind. 

In the supplied screenshots first you see a part of the original data, then the mask created by extracting by attributes through the minimum mapping unit, so areas larger than the MMU are masked. Finally a screenshot of the current result where the unmasked areas are filled in with no regard for the original boundaries of the < MMU areas.

I can't find how to do this, does anyone know a way?

P.S./on a side note

If anyone has suggestions for methods of generalizing classified raster images I'd be very interested as well!

[DanPatterson_Retired]

selective nibble isn't an option, sometimes, a Shrink followed by an Expand will get rid of small zones, but a guarantee of retaining original boundaries wouldn't be possible.  If you have a polygon file of the original boundaries, or you can make one from the raster, you might have some success with Dissolve

[thomaswarnaar]

Thank you for your reply Dan. 

I was thinking about converting the raster to polygons for exactly this reason. However the total area is the size of a medium-small country which might give problems and certainly will lead to very long computing times I think. Also part of the data that the end product is derived from was supplied in vector format (which I converted to raster) so I'm worried about the accuracy resulting from this back and forth conversion, no idea whether these concerns are at all warranted though. If all else fails I'll try the back to polygon method. 

However even then how would I approach the addition of the smaller polygons to logically selected larger polygons automatically?