Letís discuss the main, fundamental recommendations that help
to use LESSA properly. How to analyze LESSA results - extreme regions,
high gradient zones, elongation lines, long lineaments - is described
in the South-West Africa image analysis example.
Particular methods for certain investigations one can find in the
automatic lineament analysis applications
Information provided by LESSA, and first of all rose-diagrams and
fields, is objective image (texture) description. Only a specialist
can decide if the image reflects those structures that he is looking
for or not. But to obtain suitable characteristics of the area itís
necessary to choose proper input data.
First of all the image must be clean, not covered with clouds,
artificial objects, noise etc. Notice that accurate processing of
Ukrainian and Hungarian images where land is covered with agricultural
fields gave appropriate structural information.
The second important decision that user must make is the input
image scale (pixel size) choosing. We must notice that this choice
is quite different from the traditional scale choice for the visual
image analysis. We can be guided by the size of the basic structures,
those that "born" stripes - the basis of the LESSA processing.
According to their size one can choose the image scale (digitization
step or pixel size). Itís necessary to take into account that different
scale can give us not the same image description with more details
or less, but we can get description of the OTHER image, that is
formed by the structures of the other size, and can be rather different.
Itís obvious that if too coarse resolution is used we can miss the
basic structures and if the resolution is too high we can detect
small details of those structures with different orientation.
Rose-diagrams sliding window size determines size of the structures
that could be detected by the stripes statistical analysis. We must
notice that the window canít be of arbitrary size. Statistical reasons
give the working size range. The optimal size is about 64 pixels.
In the much larger windows rose-diagrams appear to be almost round,
without essential distinctions. In the smaller window there are
too few stripes, so the roses shape is coarse and its descriptors
are not stable.
Long lineaments detection also needs such accurate resolution choice
- if the resolution is too high lineaments can turn out to be meandering,
not straight and LESSA will miss them.
There is also specific scale choice for schemes and DTM.