only at the corners. For example, the full resolution class map for the southern
mosaic had 1,541,906 raster polygons ranging in size from a single pixel to over
100,000 pixels. Both accuracy assessment and vector conversion of the thematic
map is greatly hampered by so many polygons. The second step is to define a
MMU. For this project, a MMU of 25 pixels (225 m2) was selected. At 9 m2 per
pixel, 25 pixels cover only 0.0225 ha (0.0556 acres). The next step is to delete all
raster polygons that have fewer than 25 pixels. This operation is synonymous
with applying a polygon sieve to the thematic image, where all raster polygons
below the MMU threshold size are "dropped." The final step in the technique is
to iteratively apply a majority filter to the sieved image. This majority filter,
usually a 33 pixel matrix, uses a neighborhood operation to reassign sieved
pixels to a larger polygon. As the center pixel passes over each dropped pixel, the
class values for the eight surrounding pixels in the matrix are tallied. The center
pixel is then assigned the same class value as the majority of the pixels in the
matrix. The filter then moves to the next sieved pixel and repeats the operation.
The filter is passed over the image many times until all pixels within the deleted
polygons are reassigned a new class value. The operation ignores any
background pixels that were already present in the full resolution thematic map.
The threshold of 25 pixels was assumed to be adequate for retaining the ma-
jority of the thematic information throughout the study site, while removing a
significant amount of thematic speckle (or noise) and reducing the number of
raster polygons to 32,710.