Gestalt psychology and visual composition: proximity, similarity, closure, and the laws of perceptual organization

Intuition Beginner

Why do you see a "triangle" when three dotted lines almost, but do not quite, meet? Why does a grid of evenly spaced dots look like rows one moment and columns the next? Your brain is not reading the dots as a flat list. It is organising them into shapes, lines, and groups before you can stop it. This automatic organisation is what the German word Gestalt names — a shape, a form, a configuration that hangs together.

Max Wertheimer, working in Frankfurt in 1912, gave the first laboratory demonstration of this with apparent motion: two stationary lines flashed in quick succession are seen as a single line moving between them, even though nothing actually moves. He called the effect phi. From it he drew the central claim of Gestalt psychology: the whole is different from the sum of its parts. You cannot reconstruct the experience of motion by adding up two stationary flashes.

The same principle governs how you read a painting, a poster, or a web page. Nearby marks group together; similar marks group together; gaps are silently closed; one region jumps forward as the subject while another sinks back as the background. These are the laws of perceptual organisation, and they are why visual composition works at all.

Visual Beginner

Law	What it does	A one-line example
Proximity	Nearby elements group	12 dots with a gap after the 6th read as two groups of 6
Similarity	Like elements group	Red dots group apart from blue dots of equal spacing
Closure	Gaps are filled	Three broken arcs are seen as one triangle
Good continuation	Lines keep going	Two crossing curves are seen as two curves, not four segments
Figure-ground	One region jumps forward	The Rubin vase flips between vase and two faces
Common fate	Co-moving elements group	Three arrows pointing right group against three pointing left
Prägnanz	The simplest organisation wins	An ambiguous shape resolves to its most regular reading

Worked example Beginner

The Rubin vase, published by the Danish psychologist Edgar Rubin in 1915, is a single black-and-white drawing. It contains exactly two regions — black and white — separated by one shared contour. Yet it yields two completely different percepts: a white vase standing on a black background, or two black faces in profile facing each other across a white background.

The crucial fact is that you can hold only one of these percepts at a time. The vase and the faces never appear together. The moment one region becomes the figure — the thing with shape, edges, and a nearness to you — the other collapses into the ground, a formless field that simply recedes.

Step 1. Count the regions: there are 2 (one black, one white).

Step 2. Count the shared boundary: there is 1 closed contour that separates them.

Step 3. Count the stable percepts: there are 2 (vase, or two faces), and they alternate.

Step 4. Time the alternation: in laboratory measurements the spontaneous flip between the two percepts occurs every few seconds, and you cannot, by effort alone, hold either one for long.

What this tells us: perception is an active construction, not a passive receipt of light. The drawing has not changed, yet what you see changes — because your visual system must choose, at every instant, which region is figure and which is ground.

Check your understanding Beginner

Formal definition Intermediate+

Let $S=\{s_1,\dots ,s_n\}$ be a finite set of stimulus tokens in the visual field (dots, line segments, coloured patches). A perceptual grouping of $S$ is a partition $\mathcal{G}=\{G_1,\dots ,G_k\}$ of $S$ into non-empty blocks. The Gestalt laws are criteria that select one partition ${\mathcal{G}}^{\ast }$ from the set of all possible partitions, given the geometry and attributes of the tokens.

Proximity. Define a distance $d(s_i,s_j)$ on $S$ (Euclidean distance between token centroids). The proximity principle selects the partition minimising within-block distances relative to between-block distances. Concretely, for a one-dimensional array of equally spaced tokens at spacing $a$, the introduction of a single gap of size $g>a$ partitions the array into two blocks if and only if $g/a$ exceeds a perceptual threshold (empirically near $1.0$ to $1.3$ for short arrays, rising with row length).

Similarity. Let each token carry an attribute vector $a(s_i)$ (colour, size, orientation). The similarity principle groups tokens whose attribute vectors are close under some metric. The joint operation of proximity and similarity selects the partition that is stable under simultaneous perturbation of both position and attribute.

Closure. Given a set of contour fragments, the closure principle completes a closed curve across gaps whenever the completed curve has low total curvature and the gap is small relative to the fragment lengths. The perceived shape is the completed whole, not the sum of the fragments.

Good continuation. Two contour segments meeting at a point are perceived as a single continuing contour if their tangent directions agree at the meeting point (within a tolerance). At a crossing of two curves, the visual system binds the two segments whose tangents are collinear, yielding two perceived curves rather than four.

Figure-ground. A display partitions the visual field into regions. The figure-ground principle assigns to each region one of two roles: figure (bounded, shaped, nearer, memorable) or ground (unbounded, formless, farther). The assignment is a function of cues including surroundedness, convexity, size, and contrast; it is stable under perturbation of the cues and is, in the Rubin case, bistable.

Common fate. Tokens that move with the same velocity vector group together, independently of their spatial separation. This is the temporal analogue of proximity.

Prägnanz (good figure). Among all partitions compatible with the stimulus, the perceived partition is the one with the smallest structural complexity — the simplest, most regular, most stable organisation. Prägnanz is the unifying principle from which the others are derived as special cases.

The founding theoretical claim is the anti-structuralist thesis of Wertheimer, Köhler, and Koffka: the whole is different from the sum of its parts. A perceptual whole has properties — the motion in phi, the shape in closure, the figure-ground asymmetry — that no individual token possesses and that cannot be recovered by any additive computation over the tokens.

Counterexamples to common slips

• Slip: "Gestalt means the whole is greater than the sum of its parts." The canonical formulation is different from, not greater than. The thesis is qualitative, not quantitative: the whole has properties no part has, not merely more of some quantity the parts have.

• Slip: "The Gestalt laws are heuristic rules of thumb." In the computational reformulation of Desolneux, Moisan, and Morel (2008), the laws become decision procedures under an explicit probabilistic background model. A grouping is perceived when the probability of its arising by chance, against the background model, falls below a threshold (the Helmholtz or a-contrario principle). The laws are statistical detection rules, not rules of thumb.

• Slip: "Figure and ground are symmetric." They are not. Rubin (1915) established the asymmetry: the figure has shape, the ground does not; the figure is nearer, the ground farther; the figure is remembered, the ground forgotten. The Rubin vase is bistable, but at any instant the asymmetry is strict.

• Slip: "Closure means the brain draws missing lines." Closure is the perception of a completed shape, not the inference of a particular contour. The Kanizsa triangle (Kanizsa 1955) is perceived as a solid white triangle sitting above three discs, even though no triangle contour is present in the stimulus; the brain represents the figure, not a specific pixel-level contour.

Key theorem with argument Intermediate+

Theorem (Prägnanz as the unifying grouping principle). The grouping effects attributed separately to proximity, similarity, closure, and good continuation are not four independent laws but four manifestations of a single economy principle. Formally, let $\Pi (S)$ be the set of all partitions of the stimulus $S$, and let $C(\mathcal{G})$ be a structural-complexity functional assigning to each partition a non-negative cost that is small when the blocks of $\mathcal{G}$ are spatially compact, attribute-homogeneous, and bounded by low-curvature closed contours. Then the perceived partition is

$${\mathcal{G}}^{\ast }=\arg \underset{\mathcal{G}\in \Pi (S)}{\min }C(\mathcal{G}),$$

and the named laws are recovered as the conditions under which the minimiser is determined by a single term of $C$.

Argument. The claim is established by showing that each named law is the special case in which one term of the cost functional dominates, and that the minimiser is stable under perturbation of the stimulus up to a critical ratio — the threshold phenomena observed experimentally.

(1) Proximity as the spatial term. Take $S$ to be a collinear array of $n$ identical tokens at positions $x_1<\cdots <x_n$, so the attribute term of $C$ is constant across partitions and only the spatial term matters. Let the spacing be $a$ between consecutive tokens except at one location, where it is $g$. A two-block partition splitting at the gap has cost lower than the one-block partition by an amount proportional to $(g-a)$, and lower than any finer partition because finer splits break compact sub-arrays. Hence the minimiser is the two-block partition exactly when $g>a$, with a perceptual threshold accounting for noise; this is the proximity grouping. The threshold $g/a\approx 1.0$ to $1.3$ observed in Wertheimer's 1923 experiments is the empirical value at which the spatial term decisively selects the two-block minimiser.

(2) Similarity as the attribute term. Fix the positions so the spatial term is constant and vary the attribute vectors. The minimiser of the attribute term is the partition into attribute-homogeneous blocks; a checkerboard of red and blue dots at fixed positions groups by colour rather than by row, because the attribute term of $C$ dominates. When the spatial and attribute terms conflict (interleaved red and blue dots at uniform spacing), the perceived grouping depends on the relative weights of the two terms, and the system is multistable.

(3) Closure and good continuation as the contour term. Replace the token set by a set of contour fragments. Define the contour term of $C$ to be the total curvature of the completed contour, summed over completions. A set of three arcs, each subtending a gap small relative to its length, admits a completion into a closed curve of low curvature (a triangle); any alternative binding of the fragments into contours yields higher curvature. The minimiser is the closed triangle, recovering closure. At a crossing of two smooth curves, the binding that pairs collinear segments yields two contours of low total curvature, against three or four higher-curvature alternatives; this recovers good continuation.

(4) Stability and the threshold structure. Because $C$ is a sum of terms each continuous in the stimulus parameters, the minimiser ${\mathcal{G}}^{\ast }$ is locally constant on the stimulus space. It changes only when two candidate partitions exchange the minimum — that is, along surfaces where the costs are equal. Crossing such a surface produces the perceptual flip observed in the dot-grid, the Necker cube, and the Rubin vase. The threshold phenomena are thus a direct consequence of the minimisation structure, not an additional empirical fact.

(5) Recovery of the named laws. Proximity is the spatial term, similarity the attribute term, closure and good continuation the contour term, and common fate the temporal term (grouping by shared velocity). Prägnanz is the assertion that the perceived partition is the global minimiser of the combined cost. The named laws are not independent principles but the terms of a single functional.

The formulation is consistent with the computational reformulation of Desolneux, Moisan, and Morel ^{[DesolneuxMoisanMorel2008]}, in which a grouping is perceived when its probability under a naive background model falls below a threshold. The Helmholtz principle and the complexity-minimisation principle select the same partitions in the regimes where both are defined, and each supplies what the other lacks: the complexity principle gives a deterministic account of the perceived organisation, the Helmholtz principle gives the statistical criterion under which that organisation is unlikely to be accidental.

$\square$

Bridge. The Prägnanz-as-minimisation thesis builds toward 29.03.04, where the orientation- and ocular-dominance architecture of the primary visual cortex supplies the neural substrate that any grouping computation must run on, and the thesis appears again in 34.06.03, where the Bauhaus pedagogy of Itten and Albers operationalises the same principles as design rules. The foundational reason is that grouping is a single optimisation seen under different decompositions of its cost, and this is exactly the central insight that generalises from the static dot grids of Wertheimer to the dynamic, motion-segmented displays of modern vision science. Putting these together, the bridge is between the 1923 perceptual grouping laws and the 2008 computational detection framework: both identify the perceived partition with the outcome of a single selection procedure, and the pattern recurs in predictive-coding accounts 34.03.03 where the brain's top-down model and the bottom-up stimulus negotiate exactly such a minimisation at every instant.

Exercises Intermediate+

Developments and applications Master

Seven named developments carry the Gestalt programme from its Frankfurt founding to its contemporary computational and neuroscientific reformulations.

Development 1 — Wertheimer and the phi phenomenon (Frankfurt, 1912). Max Wertheimer's experiments on apparent motion, conducted in the Frankfurt laboratory of Friedrich Schumann with the assistance of Kurt Koffka and Wolfgang Köhler, established that two stationary lines presented in succession at a suitable temporal interval (roughly $60$ to $200$ milliseconds) are perceived as a single line moving smoothly between the two positions. The perception is of pure motion: at the optimal interval, the observer sees motion without seeing either line as located at its endpoint. Wertheimer termed the effect phi and reported it in "Experimentelle Studien über das Sehen von Bewegung" (Zeitschrift für Psychologie 61, 1912) ^{[Wertheimer1912]}. The result is the founding experimental fact of Gestalt psychology, because the motion is a property of the whole two-flash display that no single flash possesses: it cannot be recovered from the parts.

Development 2 — Wertheimer's grouping principles (Berlin, 1923). Having moved to Berlin, Wertheimer published "Untersuchungen zur Lehre von der Gestalt II" (Psychologische Forschung 4, 1923, 301-350) ^{[Wertheimer1923]}, the systematic enumeration of the grouping principles. The paper introduces the dot-grid experiments that demonstrate proximity grouping, the similarity experiments with varying token attributes, the common-fate experiments with co-moving tokens, and the good-continuation analysis of contour binding at crossings. The 1923 paper is the canonical primary source for the grouping laws as taught today; its experimental method (controlled variation of a single stimulus parameter with verbal report of the perceived grouping) became the standard paradigm of Gestalt experimental psychology.

Development 3 — Köhler's systematic programme (1929) and the isomorphism principle. Wolfgang Köhler's Gestalt Psychology (Liveright, 1929) ^[Kohler1929] systematised the programme for an English-speaking audience and advanced the isomorphism principle: that the perceived organisation is structurally isomorphic to the underlying brain process, so that the Gestalt properties of the percept correspond to Gestalt properties of cortical activity. Köhler's electrophysiological work on figure-ground, reported in Dynamics in Psychology (1940), pursued this principle through direct measurement of cortical fields; the strong form of the isomorphism claim (that perception is literally a property of continuous cortical field activity) was not sustained by later neurophysiology, but the weaker methodological form — that perceptual organisation must be grounded in neural organisation — became a foundational assumption of cognitive neuroscience.

Development 4 — Koffka's synthesis and the application to art (1935; Arnheim 1954). Kurt Koffka's Principles of Gestalt Psychology (Harcourt Brace, 1935) ^[Koffka1935] is the canonical English-language systematic treatise, integrating the grouping principles, figure-ground theory, the constancies (size, shape, colour), and the field theory of perception into a single framework. Rudolf Arnheim's Art and Visual Perception: A Psychology of the Creative Eye (University of California Press, 1954; revised edition 1974) ^{[Arnheim1954]} applied the Koffka framework to pictorial composition, arguing that the laws of perceptual organisation are the laws by which visual art achieves its expressive effects. Arnheim's treatment of balance, shape, form, growth, space, light, colour, movement, and dynamics as applications of Gestalt principles established the dominant framework for the psychological analysis of art through the late twentieth century.

Development 5 — The emigration to the New School and the American reception (1933-1945). The rise of National Socialism in 1933 dispersed the Berlin and Frankfurt Gestalt schools. Max Wertheimer, dismissed from his Frankfurt chair, emigrated to the New School for Social Research in New York in 1933 (the New School's University in Exile, founded the same year by Alvin Johnson, became the institutional home of the émigré Gestalt psychologists). Köhler, who had publicly opposed the Nazi regime within the University of Berlin, emigrated to Swarthmore College in 1935. Koffka had already moved to Smith College in 1927. The American reception, mediated through Koffka's 1935 treatise and the émigré teaching at the New School, carried the Gestalt programme into the postwar American psychology curriculum and, through Arnheim's 1954 book, into the design and art-history curricula that shaped the Bauhaus diaspora's second generation.

Development 6 — Computational Gestalt (Desolneux, Moisan, Morel, 2008). The mathematical formalisation of the Gestalt laws as statistical detection rules was accomplished by Agnès Desolneux, Lionel Moisan, and Jean-Michel Morel in From Gestalt Theory to Image Analysis: A Probabilistic Approach (Springer, 2008) ^{[DesolneuxMoisanMorel2008]}. The framework operationalises the Helmholtz principle (that we perceive the groupings least likely to arise by accident) as an a-contrario hypothesis test: a grouping is detected when the expected number of equally or more structured configurations under a naive background model falls below a threshold (the Number of False Alarms). The framework yields algorithmic detectors for alignments, clusters, vanishing points, and other Gestalt groupings directly from the image statistics, with a false-alarm rate controlled by the threshold. This is the modern mathematical reformulation of Prägnanz: the perceived partition is the one least probable under noise.

Development 7 — Predictive coding and the neural reformulation (Clark 2013). The reformulation of perception as hierarchical predictive coding, advanced by Andy Clark in "Whatever next? Predictive brains, situated agents, and the future of cognitive science" (Behavioral and Brain Sciences 36, 2013, 181-203) ^[Clark2013], recasts the Gestalt laws as consequences of the brain's top-down generative model. Under this account, the brain continually predicts its sensory input from a hierarchical generative model and perceives the residual prediction error; the Gestalt groupings emerge because the model's priors privilege compact objects, smooth contours, and stable figure-ground assignments. The predictive-coding reconstruction unifies the Gestalt programme with the Bayesian brain hypothesis and supplies a neural mechanism (hierarchical prediction-error minimisation) for the perceptual organisation that Wertheimer described functionally. The neural substrate of the grouping computation, in the orientation-tuned cells and horizontal connections of the primary visual cortex established by Hubel and Wiesel, provides the implementation level.

Synthesis. The seven developments fit together as a single trajectory in which the central insight is that perceptual organisation is an active selection of the simplest stable partition of the stimulus, and each development both builds toward and appears again in every later one. The foundational reason is that the phi phenomenon (Development 1) and the grouping laws (Development 2) establish the same fact at different scales: the whole has properties no part has, and the perceptual system selects a single organisation from the set of compatible ones. This is exactly the structure that Köhler's isomorphism principle (Development 3) grounds in neural activity and that Koffka's synthesis (Development 4) extends to the full perceptual field; the bridge is between the 1912 laboratory demonstration and the 1954 application to pictorial art, where the same grouping laws govern both the dot grid and the painted canvas. Putting these together, the pattern generalises from the American reception (Development 5) through the computational reformulation (Development 6), where the Helmholtz principle identifies the perceived partition with the one least probable under noise, to the predictive-coding account (Development 7), which identifies it with the highest-probability hypothesis under the brain's generative model. The pattern recurs because the selection problem is invariant under change of formalism: whether the cost functional is descriptive complexity (Prägnanz), expected false alarms (Desolneux-Moisan-Morel), or negative log posterior (predictive coding), the minimiser is the same partition, and the grouping laws are the conditions under which each term of the functional is decisive.

Full argument set Master

Proposition (Grouping stability and the threshold structure of perceptual flips). Let $S$ be a stimulus configuration and let $C:\Pi (S)\to {\mathbb{R}}_{\ge 0}$ be a structural-complexity cost functional on the partitions of $S$, continuous in the stimulus parameters. Suppose $C$ is a finite sum of terms $C=C_{\text{spatial}}+C_{\text{attr}}+C_{\text{contour}}+C_{\text{temporal}}$ corresponding to proximity, similarity, closure-and-continuation, and common fate. Then the perceived partition ${\mathcal{G}}^{\ast }=\arg {\min }_{\mathcal{G}}C(\mathcal{G})$ is locally constant on the stimulus-parameter space and changes only along codimension-one surfaces on which two candidate partitions exchange the minimum. In particular, every perceptual flip (as in the dot grid, the Necker cube, or the Rubin vase) occurs at a parameter value at which the costs of two partitions are exactly equal.

Argument. Let $\theta \in {\mathbb{R}}^m$ denote the vector of stimulus parameters (token positions, attribute values, contour geometries, motion vectors), so that the cost functional is a continuous function $C(\mathcal{G},\theta )$ of both the partition and the parameters. For each $\theta$, the perceived partition is ${\mathcal{G}}^{\ast }(\theta )\in \arg {\min }_{\mathcal{G}}C(\mathcal{G},\theta )$, chosen by a deterministic tie-breaking rule when the argmin is not unique.

(1) Local constancy off the tie set. Fix ${\theta }_0$ and suppose ${\mathcal{G}}^{\ast }({\theta }_0)$ is the unique minimiser, so that $C({\mathcal{G}}^{\ast },{\theta }_0)<C({\mathcal{G}}^{\prime },{\theta }_0)$ for every competing partition ${\mathcal{G}}^{\prime }$. By continuity of $C$ in $\theta$, the strict inequality is preserved in an open neighbourhood of ${\theta }_0$, on which ${\mathcal{G}}^{\ast }$ remains the unique minimiser and hence the perceived partition. The perceived partition is therefore locally constant.

(2) Characterisation of the flip locus. Define the tie set $T_{\mathcal{G},{\mathcal{G}}^{\prime }}=\{\theta :C(\mathcal{G},\theta )=C({\mathcal{G}}^{\prime },\theta )\}$ for each pair of distinct partitions. A flip — a discontinuity of ${\mathcal{G}}^{\ast }(\theta )$ — can occur only at $\theta \in {\bigcup }_{{\mathcal{G}}^{\prime }}{T}_{{\mathcal{G}}^{\ast },{\mathcal{G}}^{\prime }}$, since off this union the minimiser is locally unique by step (1). Each $T_{\mathcal{G},{\mathcal{G}}^{\prime }}$ is the zero set of the continuous function $C(\mathcal{G},\theta )-C({\mathcal{G}}^{\prime },\theta )$, hence closed; generically (when the gradient of the difference does not vanish) it is a codimension-one submanifold of the parameter space. The flip locus is the union of these submanifolds.

(3) Application to the named laws. Each named Gestalt law corresponds to a regime in which one term of $C$ dominates the others. In the proximity regime (identical tokens, varying spacing), only $C_{\text{spatial}}$ varies with $\theta$, and the flip locus is the single surface $g/a=$ threshold, where $g$ is the gap size and $a$ the base spacing — exactly the Wertheimer 1923 dot-grid result. In the figure-ground regime (the Rubin vase), the two candidate figure-ground assignments exchange the minimum along a surface in the cue space (convexity, size, contrast), and the balanced case is bistable because both assignments lie on the flip locus simultaneously.

(4) Conclusion. Every perceptual flip occurs at an exact cost-tie, and the threshold structure of the Gestalt laws — the empirical fact that groupings persist under perturbation up to a critical ratio and then flip — is a direct consequence of the minimisation structure of $C$, not an additional empirical postulate.

$\square$

Proposition (Equivalence of Prägnanz and the a-contrario criterion in the regime of common definition). In the regime where the structural-complexity cost $C(\mathcal{G})$ and the background probability model $P_0$ are both defined, the partition minimising $C$ coincides with the partition minimising the expected number of false alarms $\mathrm{NFA}(\mathcal{G})=-{\log }_2{P}_0(\text{at least one configuration at least as structured as }\mathcal{G})$, up to the choice of units.

Argument. Under the naive background model, the probability of a configuration at least as structured as $\mathcal{G}$ decreases monotonically with the structural complexity $C(\mathcal{G})$, because more structured configurations occupy smaller volumes in the configuration space and hence have lower probability under any absolutely continuous background measure. The relation is, in the large-deviation regime, of the form $P_0(\text{at least as structured as }\mathcal{G})\approx \exp (-\alpha \cdot C(\mathcal{G}))$ for a constant $\alpha >0$ depending on the background model. Taking logarithms, $\mathrm{NFA}(\mathcal{G})\approx \alpha \cdot C(\mathcal{G})+\text{const}$, so the partition minimising $C$ coincides with the partition maximising NFA (that is, minimising the false-alarm probability). The Prägnanz principle (minimise complexity) and the Helmholtz principle (maximise surprise against the background) therefore select the same partition in the regime where both are defined. $\square$

Connections Master

• Visual art: elements, principles, and composition 34.03.01. This unit is the perceptual-psychology depth specialisation of the composition half of the visual-art survey 34.03.01, which introduces the principles of design (balance, rhythm, emphasis, unity) in a single overview paragraph. The Gestalt laws supply the perceptual mechanism that those principles exploit: balance is felt because the perceptual field seeks the simplest stable figure-ground organisation; rhythm is read because proximity and similarity group repeated elements into a single perceived series. Builds toward the colour-theory unit 34.03.02 pending, where Itten's and Albers's treatment of colour relativity is intelligible only against the figure-ground and grouping framework established here.

• The Bauhaus: Gropius, Dessau, and the modernist synthesis 34.06.03. The Bauhaus (1919-1933) and the Berlin-Frankfurt Gestalt school were parallel and mutually aware responses to the same formal and perceptual problems of the Weimar period. Johannes Itten's Vorkurs (1919 onward) and Josef Albers's material and colour courses operationalise, in design pedagogy, the same anti-reductionist commitment to the priority of the whole that Wertheimer, Köhler, and Koffka defend in perception theory. The connection is not a one-way application of a scientific theory to a design practice but a convergence within a shared intellectual culture, reinforced after 1933 by the common emigration of Gestalt psychologists (to the New School) and Bauhaus masters (to Harvard, Black Mountain, Yale, the Institute of Design in Chicago) to the United States.

• Hubel and Wiesel's visual cortex architecture 29.03.04. The Gestalt programme postulated perceptual organisation at the functional level (1912-1935) a quarter-century before the neural substrate was identified. David Hubel and Torsten Wiesel's 1959-1962 discovery of orientation-tuned simple and complex cells in the primary visual cortex of the cat, and the subsequent mapping of orientation columns, ocular-dominance columns, and hypercolumns, supplied the neural machinery on which any grouping computation must run. The horizontal intracortical connections linking cells of similar orientation preference, reported in the 1980s and 1990s, are the candidate substrate for the long-range interactions that good-continuation and proximity grouping require. This unit provides the functional theory that the Hubel-Wiesel neural architecture implements; the two together constitute the perceptual-organisation problem at its two necessary levels.

• Italian Renaissance art: from Giotto to Michelangelo 34.04.03. The composition techniques of the Italian Renaissance — Alberti's 1435 construction of the picture plane as a window, the systematic deployment of one-point linear perspective (Brunelleschi c. 1415, codified by Alberti), and the organisation of figures into stable pyramidal groupings — are, from the standpoint of this unit, sustained exploitations of the Gestalt laws three centuries before those laws were named. Alberti's istoria requires that the figures compose into a single perceived whole, an explicit statement of the Prägnanz requirement; the perspectival convergence lines exploit good continuation to bind the depicted space into a unified field. The Renaissance unit supplies the art-historical depth in which the Gestalt-analytic reading of this unit is grounded; conversely, this unit supplies the perceptual vocabulary in which the Renaissance achievements are most precisely described.

Historical & philosophical context Master

Gestalt psychology was founded in Frankfurt in the years 1910-1912 by Max Wertheimer (1880-1943), working with Kurt Koffka (1886-1941) and Wolfgang Köhler (1887-1967) as observers and collaborators in the laboratory of Friedrich Schumann. Wertheimer's experiments on apparent motion, reported in "Experimentelle Studien über das Sehen von Bewegung" (Zeitschrift für Psychologie 61, 1912, pp. 161-265) ^{[Wertheimer1912]}, established the phi phenomenon and, with it, the anti-structuralist thesis that perception is not built up from atomic sensations. The systematic enumeration of the grouping principles followed in "Untersuchungen zur Lehre von der Gestalt II" (Psychologische Forschung 4, 1923, pp. 301-350) ^{[Wertheimer1923]}, after Wertheimer's move to Berlin; Köhler's Gestalt Psychology (Liveright, 1929) ^[Kohler1929] and Koffka's Principles of Gestalt Psychology (Harcourt Brace, 1935) ^[Koffka1935] are the canonical systematic treatises. The figure-ground asymmetry was established independently by Edgar Rubin in Synsoplevede Figurer (Gyldendalske, 1915) ^[Rubin1915], whose reversible vase became the field's emblematic demonstration.

The application of Gestalt theory to the visual arts is due to Rudolf Arnheim (1904-2007), whose Art and Visual Perception: A Psychology of the Creative Eye (University of California Press, 1954; revised edition 1974) ^{[Arnheim1954]} argued that the laws of perceptual organisation are the laws by which visual art achieves its expressive effects. The dispersal of the German Gestalt schools by National Socialism after 1933 carried the programme to the United States: Wertheimer emigrated to the New School for Social Research in 1933 (within Alvin Johnson's University in Exile), Köhler to Swarthmore College in 1935, Koffka to Smith College from 1927. The modern computational reformulation is due to Agnès Desolneux, Lionel Moisan, and Jean-Michel Morel, From Gestalt Theory to Image Analysis: A Probabilistic Approach (Springer, 2008) ^{[DesolneuxMoisanMorel2008]}, which formalises the grouping laws as a-contrario detection rules; the neural and Bayesian reformulation is due to Andy Clark, "Whatever next? Predictive brains, situated agents, and the future of cognitive science" (Behavioral and Brain Sciences 36, 2013, pp. 181-203) ^[Clark2013], which recasts Gestalt organisation as hierarchical prediction-error minimisation.

Bibliography Master

Primary sources

• Wertheimer, Max. "Experimentelle Studien über das Sehen von Bewegung." Zeitschrift für Psychologie 61 (1912): 161-265.

• Wertheimer, Max. "Untersuchungen zur Lehre von der Gestalt II." Psychologische Forschung 4 (1923): 301-350.

• Köhler, Wolfgang. Gestalt Psychology: An Introduction to New Concepts in Modern Psychology. New York: Liveright, 1929.

• Köhler, Wolfgang. Dynamics in Psychology. New York: Liveright, 1940.

• Koffka, Kurt. Principles of Gestalt Psychology. New York: Harcourt, Brace, 1935.

• Koffka, Kurt. "Perception: An Introduction to the Gestalt-Theorie." Psychological Bulletin 19 (1922): 531-585.

• Rubin, Edgar. Synsoplevede Figurer: Studier i psykologisk Analyse I. Copenhagen: Gyldendalske Boghandel, 1915. Parts translated as "Figure and Ground" in D. C. Beardslee and M. Wertheimer, eds., Readings in Perception, pp. 194-203. Princeton: Van Nostrand, 1958.

• Kanizsa, Gaetano. "Margini quasi-percettivi in campi con stimolazione omogenea." Rivista di Psicologia 49 (1955): 7-30.

• Arnheim, Rudolf. Art and Visual Perception: A Psychology of the Creative Eye. Berkeley: University of California Press, 1954. Revised edition, 1974.

• Duncker, Karl. "Über induzierte Bewegung (Ein Beitrag zur Theorie optisch wahrgenommener Bewegungen)." Psychologische Forschung 12 (1929): 180-259.

Modern and computational sources

• Desolneux, Agnès, Lionel Moisan, and Jean-Michel Morel. From Gestalt Theory to Image Analysis: A Probabilistic Approach. Interdisciplinary Applied Mathematics, vol. 34. New York: Springer, 2008.

• Clark, Andy. "Whatever Next? Predictive Brains, Situated Agents, and the Future of Cognitive Science." Behavioral and Brain Sciences 36, no. 3 (2013): 181-204.

• Hubel, David H., and Torsten N. Wiesel. "Receptive Fields, Binocular Interaction and Functional Architecture in the Cat's Visual Cortex." Journal of Physiology 160, no. 1 (1962): 106-154.

• Gombrich, E. H. Art and Illusion: A Study in the Psychology of Pictorial Representation. London: Phaidon, 1960.

• Hochberg, Julian. Perception. Englewood Cliffs, NJ: Prentice-Hall, 1964. 2nd ed., 1978.

• Palmer, Stephen E. Vision Science: Photons to Phenomenology. Cambridge, MA: MIT Press, 1999. Esp. ch. 6 on perceptual organisation.

• Rock, Irvin. The Logic of Perception. Cambridge, MA: MIT Press, 1983.

• Wagemans, Johan, James H. Elder, Michael Kubovy, Stephen E. Palmer, Mary A. Peterson, Manish Singh, and Rüdiger von der Heydt. "A Century of Gestalt Psychology in Visual Perception: I. Perceptual Grouping and Figure-Ground Organization." Psychological Bulletin 138, no. 6 (2012): 1172-1217.