" have
|
dence).
" more
Lr
ach
1
ui
ng
itives
le con-
d.
e
don-
ticle
only
ind
lay
'e
| the
'e,
ire
'On-
| the
'iation
ne.
IS,
S
re
ment
er
er-
only
.screte
le
.he
le
Uy
sults.
gions
sults.
> tonal
r-re-
really
jeci-
["pri-
3 Or-
axpect
-s de-
-eview
+384 —
of the work done to date, we will discover that each of the
existing methods tends to emphasize one or the other aspect
and tends not to treat each aspect equally.
2.0 REVIEW OF THE LITERATURE ON TEXTURE MODELS
There have been eight statistical problems to the measurement and
characterization of image texture: autocorrelation functions, op-
tical transforms, digital transforms, textural edgeness, structural
elements, spatial gray tone co-occurrence probabilities, gray tone
run lengths, and auto-regressive models. An early review of some
of these approaches is given by Hawkins (1970). The first three
of these approaches are related in that they all measure spatial
frequency directly or indirectly. Spatial frequency is related
to texture because fine textures are rich in high spatial frequen-
cies while coarse textures are rich in low spatial frequencies.
An alternative to viewing texture as spatial frequency distri-
bution is to view texture as amount of edge per unit area. Coarse
textures have a small number of edges per unit area. Fine textures
have a high number of edges per unit area.
The structural element approach of Serra (1974) and Matheron (1967)
uses a matching procedure to detect the spatial regularity of
shapes called structural elements in a binary image. When the
structural elements themselves are single resolution cells, the
information provided by this approach is the autocorrelation
function of the binary image. By using larger and more complex
shapes, a more generalized autocorrelation can be computed.
The gray tone spatial dependence approach characterizes texture
by the co-occurrence of its gray tones. Coarse textures are those
for which the distribution changes only slightly with distaace
and fine textures are those for which the distribution changss
rapidly with distance.
The gray level run length approach characterizes coarse textures
as having many pixels in a constant gray tone run and fine tex-
tures as having few pixels in a constant gray tone run.
The auto-regressive model is a way to use linear estimates of a
pixel's gray tone given the gray tones in a neighbourhood con-
taining it in order to characterize texture. For coarse textures,
the coefficients will all be similar. For fine textures, the co-
efficients will have wide variation.
The power of the spatial frequency approach to texture is the
familiarity we have with these concepts. However, one of the
inherent problems is in regard to gray tone calibration of the
image. The procedures are not invariant under even a linear
translation of gray tone. To compensate for this, probability
quantizing can be employed. But the price paid for the invariance
of the quantized imager under monotonic gray tone transformations
is the resulting loss of gray tone precision in the quantized
image. Weszka, Dyer, and Rosenfeld (1976) compare the effective-
ness of some of these techniques for terrain classification. They
conclude that spatial frequency approaches perform significantly
poorer than the other approaches.