image might intuitively appear as an act of overprocessing, the result
(Table 1) was closer to the original than the 'IBIS' technique alone.
Computer Labeling of Manual Delineations. Automated procedures can be
used for delineating polygons on satellite data, but the photo inter
preter is likely to produce a result that needs no post-processing and
is better suited for management uses. On the other hand, the computer
can be 'trained' to apply polygon labels that are more accurate and
consistent than the photo interpreter. Thus, a procedure was developed
(Woodcock, et al., 1980) to blend the best of human skills and computer
consistency. This procedure can also double as a raster-to-vector con
version technique: 1) The polygon boundaries are prepared by the USFS
using their WRIS vector system; 2) The vector boundaries are converted
to raster format using standard VICAR/IBIS software; 3) The FOCIS
stratum map is registered to the polygon map; 4) Labels are assigned to
each polygon based on a voting rule of the number of classified pixels
falling within each polygon; and 5) Labels are transfered to WRIS in a
text format.
An important advantage of this procedure is that many US National
Forest (and other management areas) already create polygon delineation
maps, and many of them are regularly put into machine format. Thus,
there is little need for an agency to change its standard operating
procedures or develop a new computer interface. Furthermore, since
boundaries in natural environments are relatively more stable than
vegetative labels, the same boundary maps could be used repeatedly with
little updating and thus better preserve the continuity between subse
quent stratifications than is typical with raster satellite classifica
tions .
The key to the successful use of this procedure is a suitable voting
rule. Initially, a simple plurality rule was tested. This was found
wanting, and a new approach was devised. The new label decision
criterion uses separate voting procedures for each of the three FOCIS
label components. Each pixel in the stratified image contains label
distinctions for each of three characteristics: regional type, size,
and density. Under the plurality decision criterion, the most commonly
occurring three-part label was assigned to the polygon. In the new
approach, each component of the label is voted on independently of the
others and the three winning components are combined to provide the new
label. For example, consider a polygon containing pixels labeled ac
cording to the following proportions: 30% M3P, 25% M4P, 25% M4S, and
20% M4N. In this situation, the label M3P would be assigned under the
plurality voting rule. However, following close inspection, a size
class of 3 is not appropriate since 70% of the pixels received a size
class 4 label. Using the new label decision criterion, a more ap
propriate label of M4P would be assigned to the stand. The same tech
nique is applied to all three label components.
CONCLUSION
Three different alternatives in post-processing for raster-to-vector
conversion have been discussed. Selection of the technique best suited
for a particular application depends on the peculiarities of the ap
plication. All three techniques employ seme form of simplification,
but it is more apparent in the spatial filtering approach.
The Modified Davis and Peet technique best preserves the original
histogram and experiences the least amount of pixel class conversion.
However, it permits diagonal connections which are undesirable for