5.8 Photographic representation and its limits🔗⧉

A small print can suggest a sunlit valley without emitting sunlight or giving the viewer the valley's depth. It carries selected cues within a limited surface. Making a photograph means deciding which cues and relationships must survive that translation.

Durand's depiction course and NPAR discussion organize the problem around a flat, finite picture with one viewpoint, fixed time, and limited contrast and gamut Durand 2002. These constraints explain why a physically direct recording may need a different rendering to convey the intended scene. They do not imply that every alteration preserves the same information.

Durand describes depiction as “the inverse of an inverse problem” Durand 2002. Vision infers a scene from the light reaching the eyes. Making a picture asks which new pattern, constrained by paper or a display, will support the intended inference. The picture need not reproduce every physical stimulus to convey the scene. An emphasized occluding contour can clarify depth, and a glow around a lamp can suggest brightness beyond what the medium emits. These are choices about cues. They can improve the intended depiction while changing the measurement, which is why realism and literal recording must be evaluated separately.

5.8.1 Compensation, accentuation, conflict🔗⧉

Compensation strengthens a cue the picture can carry to stand in for one it cannot: a cast shadow can clarify an object's height above a surface. Accentuation uses the medium's constraints expressively: a crop can make a pattern read as a flat arrangement. Conflict deliberately gives inconsistent cues, as in impossible architecture or forced perspective.

Changing the capture or display technology is another intervention. Stereo can supply distinct views to the eyes, video can sample change, and an HDR display can emit a wider luminance range. These technologies alter specific constraints; none abolishes every limit of a recorded and displayed representation. Keep that engineering question separate from the expressive choices above.

5.8.2 The picture is flat: depth and its cues🔗⧉

An ordinary photograph carries occlusion, perspective, relative size, shading, cast shadows, texture gradients, atmospheric attenuation and recorded defocus. A fence receding into the distance gives several cues together: smaller posts, closer spacing and overlapping surfaces. Their agreement helps the viewer read depth.

The two eyes see the physical photograph from different positions, and accommodation focuses on its surface. Those signals describe the print or screen rather than the depicted distances. A viewer moving beside a static print also cannot reveal a hidden surface behind a depicted object. Stereo pairs and a recorded moving viewpoint supply different information, as the stereogram and motion-parallax demonstrations show.

The pictorial cues depend on assumptions. Familiar size requires knowing the object. Height in the frame suggests distance under a ground-plane interpretation, but fails for objects at different elevations. Shading depends jointly on shape, material and illumination. A shadow can move without its object moving. The demonstrations isolate these ambiguities; they do not establish one cue as universally stronger than another.

For a photographer, this is a reason to inspect overlaps and shadows. A small viewpoint change can separate two objects that otherwise merge into one shape. Side light can reveal a surface, while a misplaced cast shadow can confuse its position. Selective focus can direct attention, but the viewer cannot simply refocus on the blurred background.

• Occlusion (overlap): a strong depth cue, and a largely ordinal one: it tells you front-from-behind, not how far. (The overlapping animals of Lascaux already use it.) This is the cue NPAR singles out as the one to reinforce when stereo is unavailable (Figure 5.8.4⛶ fullscreen).
• Relative size: equal objects shrink with distance; a receding row of identical figures reads as a line into depth (Figure 5.8.5⛶ fullscreen).
• Familiar size: we know real-world sizes, so apparent size sets distance, and because the cue supplies a size expectation, it is the easiest to subvert (Magritte's room-filling apple; the forced-perspective tourist "holding" the leaning tower) (Figure 5.8.5⛶ fullscreen).
• Height in the visual field: an object whose base sits higher in the frame (up to the horizon) reads as farther; medieval painting, Chinese scrolls, and Bruegel's winter scenes stack figures up the field to push them back (Figure 5.8.6⛶ fullscreen).
• Linear perspective: parallel lines converge to vanishing points (Raphael's School of Athens is the textbook one-point case). This is the projective geometry of Chapter 2.2, Pinhole Image Formation and linear perspective, used as a depth cue (Figure 5.8.7⛶ fullscreen).
• Shading, and the light-from-above prior: luminance gradients model form, and the brain often reads them under a prior that light comes from above. Invert a photograph of shaded bumps and they pop into dimples; turn a crater photograph upside-down and the craters become domes; a concave plaster face mask still looks convex (the hollow-face illusion). A flat picture exploits this prior, and the reversal makes it visible (Figure 5.8.3⛶ fullscreen).
• Cast shadow: a shadow anchors an object to the ground plane and signals its height. In the Kersten–Mamassian ball-in-a-box demonstration, a single ball appears to rest on the floor or to float in mid-air depending on nothing but the position of its shadow (Kersten et al. 1997; Figure 5.8.8⛶ fullscreen).
• Texture gradient: a statistically uniform texture (pebbles, furrows, grass) compresses with distance, giving a surface its slant and recession (Constable's fields) (Figure 5.8.9⛶ fullscreen).
• Aerial (atmospheric) perspective: scattering in the air can make distant things lose contrast and take on the color of the intervening atmosphere (Leonardo's hazy blue backgrounds). Its subtler role, per NPAR, is grouping: tinting everything at one distance with a common property binds it together (Figure 5.8.10⛶ fullscreen).
• Defocus blur (depth from defocus): a photograph can bake in depth of field, throwing the background into a soft wash while the subject stays sharp, and we read the blur as depth (Chapter 2.3, Lens Image formation). But notice what it costs: the blur is the photographer's choice, frozen into the print, and it denies the viewer the ability to refocus, a manipulation we return to below, and a dimension that Chapter 31.1, Light fields 101 later hands back (Figure 5.8.11⛶ fullscreen).

Julesz's random-dot stereograms isolate binocular disparity: neither image alone contains the visible shape supplied by their relative displacements Julesz 1971. Vergence and accommodation provide additional near-distance information in natural viewing, but on an ordinary flat picture they refer to the physical surface. A pseudoscope reverses binocular disparity. When that conflicts with familiar shape, shading and occlusion, the resulting percept depends on the scene and cue reliability. The demonstration explains why disagreement between cues matters; it does not license a universal ranking in which pictorial cues always defeat stereo.

Because the pictorial cues can be controlled independently, a picture can also depict an impossible scene. Escher combines locally plausible structures into globally inconsistent architecture, and Magritte uses a familiar apple at a room-filling scale. The flat surface loses some information about a real scene while permitting depictions that no real scene could supply.

5.8.3 The picture is static: time and motion🔗⧉

A still photograph integrates a finite interval. A short integration can retain a telling pose; a long one can turn motion into a trace. The waterfall and splash in Chapter 5.2, Exposure and capture timing differ in what they say about time, even when both satisfy the photographer's intention.

Muybridge's sequences sample successive phases of action, while Marey's chronophotographs and stroboscopic images can combine phases within one picture. Painted motion lines and repeated forms represent movement through a different convention. Distinguish a measured sequence from a depiction assembled to suggest one. A composite of several instants should not be mistaken for one simultaneous event.

A sequence and a composite make different temporal claims. Muybridge's successive plates allow the viewer to inspect phases of a gallop separately Muybridge 1887. Superimposing several phases, as in chronophotography, makes their relationship visible within one frame but can obscure which things coexisted. A painted repeated limb or a motion line can suggest movement without sampling any actual trajectory. The techniques share a static display surface while differing in how time was measured and represented. That distinction matters when interpreting multiple-exposure images as evidence (Figure 5.8.14⛶ fullscreen).

A telling pose can suggest the action before and after the exposure: a skater suspended in a leap or the coiled figure of the Discobolus. Blur records a different property, the light accumulated along motion during the exposure. Velázquez's spinning wheel in Las Hilanderas uses a comparable depiction of motion in paint. Light trails turn a moving source into a path, while comic motion lines mark a path by convention. The Futurists' repeated forms, such as Balla's Dynamism of a Dog on a Leash, suggest several phases on one surface. These examples matter because the visual impression of motion does not uniquely identify how motion was measured.

5.8.4 One viewpoint, and a finite frame🔗⧉

A frame excludes information. A tight photograph can hide the surrounding crowd or remove a person whose presence changes the event's interpretation. A wider frame can retain context but make the intended subject harder to read. Neither choice is neutral with respect to every question a viewer might ask.

A mirror or a montage can add another view; a panorama extends coverage. A panorama made while rotating about one viewpoint still cannot reveal every hidden surface. Large rectilinear fields stretch peripheral objects increasingly, so panoramic projections trade straight-line preservation against other geometric properties. Choose the representation for the use, and retain the capture context when the image supports a factual claim.

The depiction lectures compare several ways of including more than one view: the mirror in Velázquez's Rokeby Venus, an engineer's plan and elevation, Hockney's photographic joiners, and Cubist depictions. Each supplies information unavailable from one ordinary projection, with different rules for how the views fit together. Behind all of them sits the tension Durand frames as what I see against what I know, often summarized by two remarks attributed to painters: Turner's that he painted what he saw, and Picasso's that he painted what he knew. Alberti's picture-window metaphor and Dürer's gridded drawing frame make the complementary choice explicit: a single framed projection commits to a particular cut through the visible world. A panorama enlarges that cut without becoming an exhaustive account of the scene.

5.8.5 The contrast is limited: dynamic range and gamut🔗⧉

Two output constraints are easy to confuse. A print cannot emit the luminance of a sunlit scene, and its ratio between darkest and lightest regions may also be smaller. Scaling all values addresses the first mismatch but not the second. For a sense of scale, a sunlit scene with deep shade can span ten thousand to one or more, while a print under room light spans a few hundred to one at best. Tone mapping, local adjustments and lighting choices affect which distinctions survive.

A bright source can be suggested by a glow or surrounding dark tones even when the display cannot reproduce its luminance. That is a perceptual representation, not an absolute photometric measurement. Likewise, gamut mapping chooses a reproducible color when the target cannot display the original one. Chapter 18.10, HDR displays and display mapping and Chapter 4.7, Synthesizing color describe the output limits; Chapter 5.5, Developing and editing a photograph shows the photographer's choices within them.

The history of depiction makes the intensity mismatch tangible. The lecture contrasts a painted sky with Brunelleschi's use of a reflecting silver surface for the sky in his perspective demonstration. A pigment and a reflection have different access to the surrounding illumination. Neither comparison can be reduced to an image's normalized pixel range.

The Zone System coordinates exposure, negative development and printing around planned tonal placements Adams, The Negative. Dodging and burning modify those placements locally. Characteristic curves compress the ends of the range, while perceptual tone-reproduction methods can also model visibility, glare or adaptation Ferwerda et al. 1996. These are distinct responses to limited output, not interchangeable names for a gamma curve. Retinex provides a related account of appearance through spatial relationships Land & McCann 1971, but it should not be read as proof that the eye simultaneously measures the entire luminance span across all environments.

5.8.6 Resolution depends on the task🔗⧉

A capture can supply enough detail for a specified print or screen without resolving everything in the scene. Cropping, a larger reproduction, small text, motion blur or a low-light capture can expose the shortfall. Adequacy for one viewing task is not evidence that resolution has been solved.

Judge the important structure at the intended output, and retain an original with useful sampling margin when future uses are uncertain. Part 14, Super-resolution and upscaling and Part 16, Deblurring address conditions under which additional measurements or priors can improve the result; they do not make arbitrary missing detail recoverable.

5.8.7 Authorship and what the photograph claims🔗⧉

The camera records light through choices of position, timing, optics and processing. Automatic capture moves some choices into software. That makes the process authored, but it does not make its relationship to the world arbitrary.

A photograph can be accurate about one property and misleading about another. A tightly cropped image may record visible positions correctly while omitting decisive context. A calibrated exposure can estimate radiance without reproducing the appearance of a bright outdoor scene on a dim display. State the claim before judging fidelity.

This is also Hertzmann's argument about the choices hidden in photography [@hertzmann-2022-choices; @hertzmann-photography-blog-2022]. Color, tone, framing and perspective are decisions in making a picture. A phone image presented as having no filter can still inherit the manufacturer's tone curve, denoising, sharpening and preferred rendering of faces. The absence of a manual edit does not remove those decisions. The apparent mechanical origin of a photograph gives it authority, which makes the decisions especially easy to overlook.

5.8.8 Faithful is not the same as realistic🔗⧉

Vision adapts to illumination and interprets surfaces. A raw measurement displayed without an appropriate rendering can look unlike the remembered scene. Color balancing or contrast compression can therefore improve a perceptual match while changing the numerical relationship to scene light.

Appearance, measurement accuracy, contextual completeness and factual authenticity are separate criteria. A pleasing face does not demonstrate faithful skin color; a metrically accurate crop does not establish a complete account of an event. The decision-tree figure records where the choices enter. It should help us name them, not excuse all choices as equivalent.

A white sheet can continue to look white under daylight and tungsten illumination although the spectra reaching the eye differ. A tilted coin can be recognized as round despite its elliptical projection. These examples distinguish an inferred property of an object from the immediate pattern of light. The depiction lecture also contrasts making a picture of what is visible with making one of what is known. A photographer participates in both: choosing a viewpoint determines the evidence, and adding fill changes the illumination of the scene being recorded.

5.8.9 Choosing the instant, choosing the viewpoint🔗⧉

The selected instant can emphasize a gesture that lasted only a fraction of a second. Edgerton's brief illumination can reveal an event too fast to inspect unaided; the event remains measured even though the photograph differs from ordinary experience.

Changing viewpoint can alter the apparent relation between people or objects. Focal length alone does not change perspective from a fixed position. The same-size portrait comparison changes camera distance as well as lens focal length. Forced perspective exploits those distance relationships, while rectification changes the projection after capture. An explanation of what was done lets a viewer distinguish those cases.

Keep a face at the same size while moving the camera farther away and increasing focal length, and its nose, cheeks and background change their relative sizes. The combined change also underlies the dolly zoom: subject size stays nearly fixed while the surrounding perspective changes. Correcting converging verticals offers another distinction between optical recording and accepted depiction. A building photographed with an upward-looking camera can be projectively correct yet look more satisfactory after rectification. Peripheral stretching in a wide rectilinear image is likewise partly a property of the projection, not necessarily a lens defect. These are reasons to explain the chosen projection rather than treating every deviation from an initial photograph as an error.

5.8.10 Time and focus select information🔗⧉

Motion rendering and focus direct attention. A blurred background can separate a subject; a brief integration can make an unstable pose appear motionless. These are conventional photographic choices whose consequences can still be described precisely. Record time, focus strategy and multi-frame processing when those facts bear on how the picture will be interpreted.

The asymmetry of selective focus is especially revealing. In the original scene, a viewer could shift attention and refocus from a near face to a distant doorway. An ordinary photograph fixes the recorded defocus: looking at the doorway does not bring its texture back. The photographer has chosen which depths remain available for inspection. Chapter 31.1, Light fields 101 describes how additional angular measurements permit a range of later refocusing choices, within the captured support.

5.8.11 Filters and lighting: building the scene's light🔗⧉

A polarizer changes reflected glare before measurement. A lamp changes the scene illumination itself. Waiting until evening changes the available illumination. All can produce a different picture of the same object, but they are different interventions.

The depiction lecture uses portrait lighting to show how illumination can carry a judgment about a person. Retain the historical examples as readings of those images rather than proofs of one universal lighting code. The same underlighting can appear threatening, theatrical or simply motivated by a lamp within the scene.

A diffusion filter redistributes light around detail; a graduated neutral-density filter attenuates one part of the field; a star filter gives bright sources structured streaks. Colored filters in black-and-white photography change the relative gray values of differently colored surfaces. Even a monochrome record therefore depends on a spectral choice.

Lighting can supply exposure, fill shadows, model shape, reveal texture and establish atmosphere. The depiction lecture's examples include Gordon Willis's overhead lighting of Don Corleone in The Godfather and Arnold Newman's portrait of Alfred Krupp in his factory. Their shadows and composition participate in the images' judgments of the subjects. Light painting makes the intervention particularly explicit: a moving lamp constructs the illumination accumulated during one exposure (Figure 5.8.20⛶ fullscreen).

5.8.12 The darkroom: tone, dodging, and burning🔗⧉

Two prints from one negative can give different weight to a sky or a face. Dodging and burning change local emphasis, and a tone curve changes contrast relationships. The negative-to-print examples make that interpretation visible. For factual uses, keep track of whether the operation merely changes visibility or hides a relevant feature; the name of the tool does not settle the question.

In a photographic print, dodging holds back exposure to make a region lighter; burning adds exposure to make it darker. The printer can use those controls to separate a face from its background, give a sky more weight, or reduce a distracting bright border. These are local decisions even when the negative is unchanged. Compare two prints of the same negative by following where the eye goes first and which details become legible, rather than asking only whether the global brightness matches. The operation changes the interpretation through visibility and emphasis (Figure 5.8.22⛶ fullscreen). The printers of the twentieth century treated this as the work itself: W. Eugene Smith reportedly spent days printing a single photograph of Albert Schweitzer, and Ansel Adams burned the sky of Clearing Winter Storm heavily to give it weight. The same negative yields as many prints as there are printers and intentions.

5.8.13 Make-up and retouching: authoring the subject🔗⧉

Styling changes the subject before capture. Retouching changes the image afterward. Brightening a face, removing a bystander and replacing a face may use adjacent controls in an editor, but their consequences for a record differ substantially.

A continuity of tools does not erase useful boundaries. A documentary claim can require that depicted people were present and that important scene elements were not removed. A calibrated measurement can permit a documented flat-field correction while excluding invented texture. An illustration can permit fabrication if its use makes that status clear. Chapter 40.2, Ethics of computational photography and Part 41, Image Forensics and Authentication examine how such claims are established and communicated.

Make-up supplies shading and highlights on the subject before the exposure. Digital contouring can alter similar cues after it. The infamous TV Guide cover combining Oprah Winfrey's head with another person's body demonstrates a further step: a photograph-like appearance assembled from people who did not pose that way together. Similarity of appearance or editing controls does not make these operations equivalent as evidence.

5.8.14 Getting closer: systematic capture and the biases that remain🔗⧉

A fixed capture protocol can reduce ad hoc choices and make a survey repeatable. It still has a sampling schedule, geographic coverage, field of view and processing policy. Evaluate whether those choices support the intended inference. An omitted location or time cannot be recovered by treating the recorded subset as exhaustive.

For a scientific or documentary record, retain source measurements and relevant metadata, document processing, and validate the particular quantities being used. These practices do not remove authorship. They make its effects inspectable and can support reliable claims despite them.

A systematic street survey replaces many individual shutter decisions with a policy. The routes decide which places are visible; the schedule decides which traffic, weather, and activities are represented. Privacy filtering can deliberately remove identifying information. Repeating the same policy makes the collection more reproducible, but does not make its coverage exhaustive or its processing interpretation-free. The appropriate question is whether the policy records the evidence needed for the proposed use, and what it systematically leaves out.

The same issue appears in datasets. A collection reflects who had access to cameras, which subjects they considered worth recording, where collection occurred and which photographs were retained. A training procedure can treat that selection as the distribution to reproduce. Underrepresented conditions then have less support, and familiar photographic conventions can be learned as though they were properties of the world. This connects photographic authorship to Chapter 7.5, Image priors: a dataset is part of the construction of a learned prior.

The history of lighter-skin reference cards in color printing is a concrete example of a reference choice becoming a technical norm. Exposure, color reproduction and detection should be evaluated across the people and conditions the system will encounter, rather than only against the examples used to tune it. Systematic capture can make choices repeatable and inspectable; repeatability alone does not establish representative coverage.