All graphs demonstrated in Figure 6 show that scale
parameter greatly impacts on image object amount and the
overall changing tendency of image object amount is down-
ward with increasing the scale parameter, no matter what
segmentation method was employed.
For the MR segmentation method, small
SP
(or low hetero-
geneity) corresponds to large amount of image object. From
the aspect of spectral heterogeneity, when shape parameter
is set small,
SP
mainly refers to spectral heterogeneity. Figure
6 (a)~(c) shows that the image object amount dramatically
decreases with increase of spectral heterogeneity. On the other
hand, from the aspect of shape heterogeneity, Figure 6(a)~(c)
also show the decline trends of image object amount for im-
age-A, B, and C are not as dramatic as those of spectral hetero-
geneity when changing shape parameter. Additionally, for the
three image A, B, and C, the spatial resolution of image A and
image B are higher and the geo-objects in these two images are
more regular than image C. It can be concluded from Figure
6(a)~(c) that when the shape of geo-objects is regular, the shape
parameter impacts the segmentation more than on the image
with lower spatial resolution or without regular geo-objects.
Scale Effect with Change of Heterogeneity Parameter
For the
MRS
method, when shape parameter is fixed as 0.1 or
shape parameter is set as small value, parameter
SP
mainly
represents spectral heterogeneity, so as to
SP
’
for
ECWS
meth-
od. For the
MSS
method, though scale parameter
M
refers to
merging threshold or the minimum pixel number of segment-
ed image object, the essence of scale effect is still spectral
heterogeneity, which is determined by the principle of
MS
clustering. Therefore, the scale effect of segmentation scale
parameter is actually manifested as the change of spectral
heterogeneity with image object amount; meanwhile the spec-
tral heterogeneity consists of intrasegment homogeneity and
intersegment heterogeneity of the segmented image object.
Intrasegment Homogeneity and
OA
with Change of Image Object Amount
From the view of overall tendency of intrasegment homogene-
ity, as demonstrated in Figure 4, with increase of image object
amount, 30 out of 33 series of experimental data strictly
represent the tendency that the intrasegment homogeneity
decreases with decrease of image object amount. However, in
spite of the volatility in the intrasegment homogeneity in the
Figure 4. Average intrasegment homogeneity and intersegment heterogeneity changing with differentparameters by using
different segmentation methods.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
November 2018
687
