------------------------------------------------------------------------------- CS 99D - Introductory lecture, January 9, 2001 Marc Levoy Stanford University (c) 2001 ------------------------------------------------------------------------------- *** The relationship of illusion to art *** Definition of Western illusionistic art: o make the picture in the frame look real, like a window o European and American art Is an illusion of reality a/the goal of art? o Yes, say most art historians from the Greek to the 19th century o Iconic before the Greeks, abstract after the 19th century Before: -> Kleitias and Ergotimos, Francois Vase (570 B.C.), Gardner, p. 133 and detail to its right After: -> Gnosis, Stag hunt (300 B.C.), mosaic, Gardner, p. 167 21- AD Pliny the Elder o only his Natural History survives o ~"Art is the imitation of nature, `mimesis'." o gave a history of illusionistic inventions 1568 Vasari o biographer of the Renaissance artists o first art historian o ~"The artist strives for mastery of representation of nature." 1843 Ruskin's Modern Painters o history of art interpreted as progress toward visual truth o ~"...imitating the image on the retina..." o Implication: the new science of perceptual psychology could eventually solve the central problem of art "Without being aware of the fact, Ruskin had thus laid the explosive charge which was to blow the academic edifice sky-high." -Gombrich A&I, p. 14 o No, says Modern (abstract) Art "It is one of the permanent gains we owe to [Modern Art] that we are rid of this type of aesthetics." - Gombrich, A&I, p. 4 o No, says photography "...science is neutral, and the artist will appeal to its findings at his peril." -Gombrich, A&I, p. 15 Artists have other goals: o narrative o emotion What then is the place of illusion in art? o tool? (in service of a narrative, which is the goal) o style? (a particular aesthetics, a taste, now passe') ------------------------------------------------------------------------------- *** The riddle of style *** The "riddle of style": -> Alain's cartoon of Egyptian art, Solso, p. 194, also in Gombrich, A&I, p. 2, What is a style? 1. A style is the current state-of-the-art in imitating nature Q. If so, then are later styles more "advanced"? o Early art is similar to children's art. -> Constable's Wivenhoe Park, Gombrich A&I, p. 304, and a child's copy, p. 292 Q. If so, then how can we leave so much out? o The beholder's share, the etc. principle -> Lorrain's The Tiber above Rome, Gombrich A&I, p. 185 Q. If so, then why do we still use line drawings? o A way to bypass early stages of perception? -> David Marr's Primal Sketch -or- 2. A style is a learned set of conventions for representing nature o E.H. Gombrich's Art and Illusion (1960) o based on 20th century perceptual psychology Gombrich says: o A style establishes a set of *shared conventions* for image-making. o Shared conventions facilitate *rapid perception*. o Shared conventions permit perception with an *economy of means.* Implications: o we draw the way we've been taught by the images we've seen "...paintings owe more to other paintings than they owe to direct observation." -Wolfflin, in Gombrich, A&I, p. 317 "Art is born of art, not of nature." -Malraux, paraphrased (?) in Gombrich, p. 24 o Example: the prevalence of formula -> Rubens's Theorie de la figure humaine (1773), Gombrich A&I, p. 165 o which lines are drawn in the lower drawing seem selected "by convention", e.g. pectoral muscle, hip Q. If so, then why do styles change over time? o they change much faster than spoken/written languages -or- 3. A style is a particular choice of representation that evokes an emotion o a non-illusionistic goal of art o allusions and metaphors o societal taste Which of these stylistic inventions are best explained by which of the above definitions of style? o chiaroscuro ("clear-obscure") o saturated shadows - Cennini's Il Libro dell'Arte (~1390) vrs o black shadows - Alberti's Della Pittura (1435-6) -> Fra Filippo Lippi's Trinity Alterpiece (1457), Lamb, p. 13, compare right and left saints, Lippi is thought to have painted the left saint o sfumato ("cloudy", "soft", "mellow"), the blurred outline o intent is to induce ambiguouity into reading of expression -> Leonardo's Mona Lisa (1502), Gombrich SoA, p. 301 and detail on p. 302 ------------------------------------------------------------------------------- Question: Setting aside non-illusionistic goals for art (attitude, emotion), and setting aside the "learned conventions" aspect of style, we finally approach a "science of illusion". Is this not solved by photography, or modern physics + computer graphics? Can we not now compute a perfect illusion? Thesis: Images are a limited medium. They cannot completely imitate reality. Special techniques are required to overcome this limitation. ------------------------------------------------------------------------------- *** Creating the illusion of reality with limited media (images) *** Depth cues: o stereo o focus ("accommodation") o head motion parallax - you move o kinetic depth effect - object moves o occlusion o linear perspective: o size of objects o placement of objects on ground plane o convergence of parallel lines o spacing of regularly spaced objects that recede in depth o orientation and scale of textures o etc. o atmospheric ("aerial") perspective: o desaturation o color shift o shading o shadows: o "attached" o "cast" Limitations of images: o monocular o constant focus - on plane of painting o fixed viewpoint - scene doesn't change if you move o static scene - objects in scene don't move o fixed perspective - appears distorted from any other viewpoint o limited color gamut o limited dynamic range Sculpture is even more limited, but Bernini describes how to "shade" a bust by changing its shape unnaturally, hence its shading under expected illumination. ------------------------------------------------------------------------------- *** The dynamic range problem in painting *** The problem of two paintings: o Helmholtz gives example of desert versus moonlit scenes o dynamic range of nature is large: 800,000:1 surface illuminated by sun : by moon (20 stops, or 1/1000 sec vrs 13 min exposure) o human contrast sensitivity: 100:1 human photoreceptors (excluding adaptation) 10:1 + variation in pupil size "very little" + photopigment bleaching 100,000:1 + neural adaptation = 100,000,000:1 - Wandell, personal communication o dynamic range of media is low: 100:1 painter's pigments (equal to human receptors without adaptation!) Cowan says: 10:1 printing 20:1 paint 100:1 CRT 200:1 transparency o Assume artists use maximum white & black pigments on both paintings. o Assume they are side-by-side in a gallery and equally illuminated. o We will have no trouble identifying them as desert or moonlit. o How is this accomplished? Artist's solution to the problem of limited dynamic range: o Our perception of a scene does not depend on *absolute* brightness. o We perceive *relative* brightness. o Thus, we can paint scenes of varying brightness using same pigments. o This is *necessary* to be able to paint a desert and moonlit scenes, o but it is not *sufficient* to distinguish these two types of scenes. o How do we distinguish them? -> Sheeler's The Upper Deck, Gardner, p. 1084 -versus- -> Wright's A Philospher Giving a Lecture at the Orrery, Gardner, p. 896 Part 1: a mathematical statement of the notion of "relative brightness" o Fechner's law: we perceive *ratios* of absolute "brightness" o the proof lies in detecting "just-noticeable differences" (JNDs) o Thus, our response to light is linear on a log scale. Part 2: response at extreme light levels o high light levels -> at saturation -> compression of contrast o low light levels -> at threshold -> expansion of contrast o In film, this would be called "reciprocity failure". Adjunct problem: side-effects of bright lights 1. physical effects, like penetration of human tissue -> glow of tissue o Solution: draw the glow -> de La Tour's The Carpenter, Solso, fig. 5.4, p. 108 (moon = candle at 12 feet - Helmholtz, p. 290) 2. perceptual effects, like scattering in the eye -> starburst patterns o Paints are not bright enough to trigger this effect o Solution: draw the starbursts -> Nakamae et al., A Lighting Model Aiming at Drive Simulators, Proc. Siggraph '90, p. 401 3. another perceptual effect, color afterimages o bright colors -> receptor fatigue -> complementary after-image o Example: red drapery -> cyan after-image in a gray area o Paints are not bright enough to trigger this effect o Solution: paint the after-image -> Seurat's La Grande Jatte (1886), Kemp, p. 333 and detail -> Delacroix's Expulsion of Heliodorus (1849-61), Kemp, p. 331, which is a detail o artists also introduce blue to reflect sky, muddying this issue -> Parramon, The Big Book of Oil painting, recipe for use of complementary colors, p. 106 o we'll come back to this issue later in the course ------------------------------------------------------------------------------- Key questions for course: 1. What do scientists know about creating an illusion of reality? o *physical* processes are well known o *low-level perceptual* processes are moderately well known o *high-level perceptual/cognitive* processes aren't well known 2. How do specific artistic techniques work to create this illusion? o physics alone o linear perspective o Euclid could verify its correctness o low-level vision o manipulation of dynamic range o Euclid & Helmholtz together could verify it o high-level perception o the shape of a shaded sphere o the Computer Vision problem o draws on universal human experience o the size of trees o the AI problem o draws on conventions of a specific culture o porphyry = imperial, ultramarine = lapis lazuli = $$ -> Gombrich SoA, p. 217 o the scholar's realm o not illusionistic, strictly a matter of taste o Impressionist's brushstrokes? o the critic's realm o the riddle of "style" 3. If we can mathematically model the limitations of images as a media, and we know and can model illusionistic techniques for overcoming these limitations, *can we "compute" optimum illusions?* o Can it be used to improve illusionistic art? o Can this be used to improve computer graphics renderings? o Even if we cannot model these effects quantitatively, can we develop human-driven (i.e. interactive) systems for creating stylistic illusions in computer graphics? 4. Can we automatically (or semi-automatically) *compute non-illusionistic stylistic renderings* e.g. for illustration of scientific data? -> Winkenbach and Salesin, Siggraph '96, p. 476 -------------------------------------------------------------------------------