Sketch-based Evaluation of
Line Filtering Algorithms
- Evaluations in GIScience
© Mahes Visvalingam, 2000

Croplogo.jpg (11317 bytes)

    The powerpoint presentation and published papers are available.from m.visvalingam@dcs.hull.ac.uk
    Some browsers do not re-display gif images.  You need to force this through a refresh/reload or by going bak and then forward again.

 

eval1.jpg (9458 bytes)
As we venture beyond approximation into caricature, we need to look at how we could evaluate competing algorithms.
  • Current mathematical evaluations based on geometry are inappropriate.
  • On the other hand, visual evaluations are too subjective.
  • Psychological experiments rely on inductions based on empirical evidence. However, this smacks too much of the philosopher's All Swans are White example. Induction is never conclusive.  Here, we do not even have tangible facts, only opinions to go on.  Mental visualisation, unlike computer visualization, is a subjective process (Visvalingam, 1994).
eval2.jpg (9475 bytes)
However, this case study shows that there is scope for conjecture-based spatial reasoning.  Unlike induction, deduction is truth preserving.   It provides us with some scope for evolving testable hypotheses based on explicitly stated assumptions, case-based evidence and logical arguments.  The kind of reasoning we could pursue is indicated below; a more detailed analysis is provided in Visvalingam and Dowson (in press).
 
 
  1. Perception of sketches assumes that the surface of terrain is continuous.
  2. The early visual system groups runs of strokes, on Gestalt principles of continuation, into linear syntactic units.   The long primitive indicating the shoulder (below A) appears to command higher priority.  This may be because long primitives which are akin to silhouettes are given higher priority than others.
  3. The angled primitive,  which appears over the top of the shoulder (to the left of A), commands a lower priority.
  4. Since the base of this primitive is in front of the shoulder, the entire primitive is perceived as being in front since it makes more sense. The primitive can only be seen as being in front if the pitch or pose of the primitive is tilted towards the user. Thus, the upper part of the primitive is not SEEN as lying on the surface although we KNOW that, by definition, all P-strokes must lie of the surface.
  5. When some marks become detached from the surface, some other marks may also appear to float.   For example, the angled primite to the upper left of the primitive.
    6.
ex2-rdp.jpg (139859 bytes)
eval3.jpg (8680 bytes)
Such inconsistencies occur because the RDP algorithm mislocates the plateau edge.  The subconscious mind perceives such inconsistency in the placement of marks.   Some people seem to ignore the output of the subconscious and appear to be focusing only on some of the layers which form the sketch.  Others have a gut feeling that 'something is wrong' but are unaware of the precise reasons, because the computation occurs within the subconscious.

If expert perception is based on knowledge-based subconsious computation, then there is some scope for basing the evaluation of sketches and of algoithms on logical consistency; conceptually, is no different to cartographers ensuring that contour interpolation is consistent with drainage patterns.    Hence, the title of the talk - sketch-based evaluation of line filtering algorithms.

This is a monumental challenge.  Even if we do not automate the computations, qualitative spatial reasoning can lead to the re-discovery of unstated principles.  This is one way in which GIS, as a science, can inform systems development.


Home/Next