PERCEPTUAL GRAPHICS FOR EFFECTIVE VISUALIZATION

View/ Open
Date
2005-08-31Author
Lee, Chang Ha
Advisor
Varshney, Amitabh
Metadata
Show full item recordAbstract
Current trends in 3D graphics point to a near future when our
ability to generate 3D content will far surpass our ability to
analyze it meaningfully. These trends have inspired us to improve
the comprehensibility of 3D graphics rendering using insights from
human perception of geometry and illumination. In this
dissertation, we develop algorithms and systems to seamlessly
integrate the low-level human visual system cues with object
modeling and lighting for 3D graphics.
Artists and illustrators have enhanced the perception of shape
with discrepant lighting for centuries. Traditional graphics
however assumes a model of consistent lighting. We have developed
a lighting design system, that by relaxing the constraint of
consistent lighting is able to convey a better depiction of shape.
Our system for automatic lighting design, Light Collages, segments
the objects into local surface patches and uses careful placement
of highlights, shadows, and silhouettes on them to enhance shape
perception. We have developed a spherical-harmonics-based
formulation to achieve a 20-fold improvement in speed.
Geometric processing in graphics has made significant advances
over the last decade in defining and using mathematical measures
of shape, such as curvature. However, less attention has been paid
to the use of perception-inspired metrics for geometric
processing. We have brought perception-inspired metrics to bear on
the problem of processing and viewing for 3D graphics. We have
developed the concept of scale-dependent mesh saliency for
graphics. We have also explored how saliency can be used to
prioritize operations in applications such as object
simplification and to automatically compute desirable viewing
parameters for 3D graphics applications.
We believe that Perceptual Graphics could lead us in the direction
of more effective graphics applications that not only use
computational resources wisely, but also lift the burden of
unnecessary rendered detail from the human visual system.