tive
ng
2100 m above ground, resulting in an average scale of 1: 7000.
In this test area, ground investigations are performed in permanent
sample areas arranged in a regular grid. From these sample areas
all those were selected in which at least 90% of all trees were of
one single species. Homogeneous sample areas with the following
tree species were observed: beech (Fagus silvatica), spruce (Picea
abies), fir (Abies alba) and pine (Pinus silvestris). The positions
of these homogeneous sample areas were marked on the photographs.
The size of the sample areas is approximately 15 m square on the
ground or 2 mm square on the photographs. For each of these
sample areas textural parameters were extracted with the Quantimet.
In addition, spectral parameters were determined by measuring the
colour densities with the aid of a densitometer Macbeth TD 504.
Textural feature extraction
The textural parameters of a sample area were obtained in the
following way: The image of the sample area was input to the Quanti -
met with the aid of a TV system combined with a microscope. Fig. la
shows an example of an input image from a sample area of pine.
The input image is then digitized in the Quantimet and converted into
a binary image by hardclipping (i.e. by introducing a grey tone
threshold and by reproducing in white all those portions of the image
that are lighter than the threshold tone, and by reproducing in black
the portions of the image that are darker than the threshold tone).
Fig. lb shows an example of a binary image obtained from the input
image.
The image processing unit of the Quantimet is then capable of
counting isolated white objects in this binary image, of determining
the area of the white objects, and of determining the perimeter of
the white objects (Fig. lc). These three parameters: number, area
and perimeter of white objects, seen as a function of the threshold
density, served as ''primary textural parameters' which were in-
vestigated in detail.
For the computation of these three parameters, it is possible to
take into account only those isolated white objects that meet certain
requirements, e.g. that exceed a given minimum size. For the
work reported here, a threshold size was defined to suppress the
granularity of the photographic film.
Fig. 2 shows the image of a sample area of beech (a) together with
binary images derived from it at various hardclipping thresholds
(b -f). In Fig. 3, the corresponding images of a sample area of