'Shape from Shading' (SFS) has been extensively studied in machine
vision (first quantitative algorithms date from the 1950's in
astronomy). Shading has also been studied as an important visual cue,
and it is of considerable importance in the visual arts. It is
remarkable that the formal theories appear not to have the slightest
affinity to generally accepted artistic usage. Moreover the bulk of the
algorithms that have been proposed are particularly awkward candidates
for biological implementation. Moreover, it is not at all evident that
the 'SFS problem' as it is generally formulated has much relevance in
real life. One aspect that is particularly doubtful concerns the
observational basis for the calculations. I show that psychophysical
evidence suggests structures that fall neatly in line with artistic
praxis. I develop a formal apparatus for 'relative SFS', a method that
seeks to find the relief with respect to some fiducial (global) shape.
Such methods are also special because they manage with local
operations, whereas conventional SFS has to rely on global algorithms.
I also reconsider the observational basis and suggest that 'surface
illuminance flow' is an observable that complements 'shading' in
realistic settings.
ABOUT THE LECTURER:
Jan J. Koenderink graduated in Physics and Mathematics in 1967 at
Utrecht University. He has been associate professor in Experimental
Psychology at the Universiteit Groningen, then in 1974 returned to the
Universiteit Utrecht where he presently holds a chair in the Department
of Physics and Astronomy. He founded the Helmholtz Instituut in which
multidisciplinary work in biology, medicine, physics and computer
science is coordinated.
He has received an honorific degree (D.Sc.) in Medicine from the
University of Leuven and is a member of the Royal Netherlands Academy
of Arts and Sciences. He participates in the editorial boards of a
number of scientific journals, scientific boards of international
conferences and scientific institutes.
Research interests include cognitive science, ecological physics and
machine intelligence.