Sketch-based evaluation of line filtering algorithms

Visvalingam, Maheswari

Abstract

CISRG research into algorithmic sketching seeks to revive interest in the lost art of landscape drawing as practised by Holmes,
Lobeck, Raisz and others. Dowson (1994) and Visvalingam and Dowson (1998) developed the P-stroke style for sketching.
Visvalingam’s algorithm for line filtering (published in Visvalingam and Whyatt, 1993) served to rank the DEM cells. This ranking
was independent of the view direction. Approximately 5% of grid cells, called core cells, lying on the major curvatures in the
terrain data were filtered using four tolerances. These tolerances were varied to suit the view and convex and concave shape of
forms. Fragments of row profiles, across the DEM, containing the core cells were filtered to provide P-stroke sketches. The
sketches consist of about a third of the height values and are thus still in the realms of minimal simplification.
Visvalingam and Williamson (1995) found that the Douglas-Peucker algorithm (1973) was better than Visvalingam's algorithm for
minimal simplification of 2D lines. Since the Douglas-Peucker algorithm is more readily available, Visvalingam and Dowson (1999)
investigated its utility for P-stoke sketching. Comparisons of filtered terrain profiles were confusing and inconclusive. When these
filtered points were plotted on contour maps, the Douglas-Peucker algorithm seemed to be better since it provided more connected
runs of core cells. However, the sketches abstracted with the Douglas-Peucker algorithm lacked Gestalt qualities of coherence and
pregnance. Whereas the P-strokes generated with Visvalingam’s algorithm were perceived to be on a single terrain surface, those
derived with the Douglas-Peucker algorithm did not. The visual system appears to have a tendency to project anomalous marks into
the foreground where they may be ignored or scrutinised. In places, these projections seem to be dependent on interpretations of
the pose of the graphic primitives, abstracted by the early visual system. The results suggest why extreme point methods for
abstracting TINs from grid DEMs may not yield a set of optimal points for plateaus.
The results indicate that P-stroke sketching provides another approach to evaluating line-filtering algorithms. While the sketches
output by the Douglas-Peucker algorithm may be instantly perceived as incoherent, the reasons for this had to be deduced. The
need for reflection and deduction suggests that the visual system, having grouped marks in the early Gestalt stages of perception,
may be performing some knowledge-based comparisons at a subconscious level. It also seems as if emotional judgements of the
aesthetic quality of drawings may be based on qualitative spatial reasoning. Variations in conceptual and procedural knowledge make
such qualitative evaluations subjective and unreliable. However, the notion that subconscious computation is triggered in a
knowledgeable mind affords a methodology for computer evaluations based on cognitive criteria. The talk will therefore focus on
sketches for cognitive evaluation of line filtering algorithms.

Citation

Visvalingam, M. (2000, October). Sketch-based evaluation of line filtering algorithms. Paper presented at GIScience 2000, Savannah, Georgia

• Presentation Conference Type: Conference Paper (unpublished)
• Conference Name: GIScience 2000
• Start Date: Oct 28, 2000
• End Date: Oct 30, 2000
• Acceptance Date: Oct 31, 2000
• Deposit Date: Oct 16, 2017
• Peer Reviewed: Not Peer Reviewed
• Keywords: Line generalisation algorithms, Terrain visualisation, Gestalt perception, Cogntiive analysis, Subconscious computation
• Public URL: https://hull-repository.worktribe.com/output/472460