Necker Cubes. Necker Cubes illustrate the claim that how one categorizes the content of an image can alter the diagnosticity of image cues, which can alter the way the input is organized into perceptual cues. Categorizing the left-hand image as a Necker Cube enables viewers to alter the relative spatial position of vertices A and B at will. Similarly, it enables viewers to perceive vertex C as either a single point in two dimensional space or as overlapping points in three-dimensional space in two different configurations. Therefore, how one categorizes a stimulus can influence how one perceives the stimulus. 

Context in source publication

Context 1

... define the shapes of objects, and the placement of objects in scenes, relative to medium-specific constraints on artistic production. Gombrich (1960, pp. 147–172) described ‘how-to’ manuals that provide instructions and simplified line drawings to train artists to the basic proportions of objects as evidence for artists’ use of schemata. For instance, in modeling a standing human figure in a dynamic pose it is critical to get the relative placement of the volume of the quadriceps and the calf muscle across the knee cap correct. The mass of the quadriceps is raised on the lower portion of the thigh in this context, and forms a roughly triangular shape whose apex points to the inside of the knee. The calf muscle is raised towards the knee and rolled to the outside of the shin. The relative downward and upward thrust of these two masses should form parallel vectors when viewed from the front (Lanteri, 1902/1985, p. 147). These types of schemata can influence perception in several ways. They direct attention to features and characteristics of an object or scene that might otherwise go unnoticed or unperceived. Further, they provide artists with frameworks for the relative placement of objects and their parts, which help locate their expected positions in the visual field (e.g., the central features of a face form an inverted equilateral triangle horizontally bisected by the midline of the head). In certain contexts the latter bias forms recognition by filtering out the effects of practical knowledge on the perception of an object (Cohen, 2005). In this section we have presented art historical perspectives that exemplify the traditional view of the influence of artists’ methods in visual analysis. Fry and Gombrich disagreed about the epistemic object of artists’ formal techniques and methods. Fry contended that artists’ methods yield knowledge of the structure of the stimulus. Gombrich argued that they represent artists’ knowledge of the structure of phenomenal experience. However, both share a critical assumption: practical knowledge functions as a bias that constrains visual analysis in ordinary perception. These biases interfere with a perceiver’s ability to recover formal cues from the visual field sufficient to induce the perception of realistic three-dimensional scenes on a two-dimensional surface. This entails that artists must develop perceptual strategies to overcome their judgments about the identities and functionally salient features of 158 W. Seeley and A. Kozbelt objects. Therefore, artists’ methods can be interpreted as sets of perceptual strategies that enhance performance in visual analysis. We propose a model for artists’ perceptual advantages (Figure 3) derived from Philippe Schyns’ (1998) diagnostic recognition framework for object identification. Theories of object identification have traditionally divided visual perception into two sets of processes: form recognition and object categorization. The standard computational argument for this division of labor rests on a logical point. Perceivers recognize the identity of objects in the visual field by matching visual representations of the geometry of a stimulus to stored records of the ordinary shapes and functions of object types. Therefore, form recognition must precede and operate independently from object categorization. However, researchers like Patrick Cavanaugh (1991), Schyns, and Stephen Kosslyn (1996) dispute this traditional computational model. They argue that object recognition does not require a full representation of the form Philosophical Psychology 159 of an object or the structure of a scene. Rather, the visual system can get by with incomplete sets of object cues that suffice to direct attention to diagnostic features in the visual field (see also Cohen, 2005, p. 999). Diagnostic features are defined as sets of image cues that are sufficient to determine the identity of a stimulus. On this account, object identification is interpreted as a hypothesis testing process. Minimal sets of image features (e.g., coarse-grained visual patterns or sets of fine-grained features) are used to develop perceptual hypotheses about the most likely identity of a target object. These hypotheses, in turn, direct attention and prime the visual system to the expectation of confirming diagnostic features at particular locations in the visual field. Therefore, Cavanaugh (1991) and Kosslyn (1996) have argued that object identification is a bootstrapping process subserved by a distributed network of different resources. Schyns’ identifies two critical variables in object recognition tasks: cue availability and cue diagnosticity. Cue availability is determined by both the perspective of a viewer and physiological and computational constraints on the availability of perceptual information to a particular system. 6 Cue diagnosticity is determined by the information processing requirements of a particular task. Consider the two geometric patterns in Figure 4. Viewers familiar with a Necker Cube readily recognize that the two interior vertices in the left-hand figure, A and B, are diagnostic for two unique cubes at different orientations. Categorizing the figure as a Necker cube 160 W. Seeley and A. Kozbelt enables viewers to alternately conceptualize vertex A or vertex B as the closest point in the foreground of the image. Viewers are subsequently able to both perceive the figure in depth and alter its perspectival projection by shifting attention from one junction to the other. The right-hand pattern is harder to perceive as a Necker Cube. One must consciously concentrate on categorizing the figure as a cube in order to perceive vertex C as overlapping A and B junctions. Otherwise it defaults to a two- dimensional pattern of six equilateral triangles. Categorizing the pattern as a cube, therefore, alters the diagnosticity of C from the depiction of a single point in a two- dimensional image to the depiction of two overlapping points separated in depth. These examples demonstrate that how one categorizes an image can alter how one analyzes a stimulus into object cues, how one organizes these cues (e.g., altering whether vertex A or B is in the foreground), and even how one perceives them (e.g., as two corners in a three-dimensional space as opposed to one point in a two- dimensional space) (Schyns, 1998, p. 150). Schyns’ diagnostic recognition framework can be used to explain the importance of specialized spatial schemata for artists’ perceptual and depictive practices. The salience of image features in a perceptual context depends on the particular perceptual hypothesis guiding attention. Artists’ spatial schemata encode knowledge of sets of image features sufficient for adequate depiction in their medium. In two- dimensional media, these sets of visual cues induce realistic three-dimensional perceptual experiences because they are image features that are diagnostic for the identity of the depicted scene or object. Therefore, schemata not only enable artists to recognize stimulus features sufficient to produce particular perceptual effects in their medium, but also confer perceptual advantages for visual analysis in ordinary perceptual contexts. Kosslyn’s model for object recognition provides a neuropsychological explanation of the role of cue diagnosticity in perception. Kosslyn argues that object recognition is an example of cooperative computation (i.e., a set of interconnected subprocesses that share information and so function as a larger distributed system) (Kosslyn, 1996, p. 121). Cooperation is a property of opportunistic processing systems (Kosslyn & Koenig, 1995, pp. 76–77). The visual system is both cooperative and opportunistic in the sense that it exploits any and all resources available to it in order to promote fast and efficient recognition of the attributes necessary for cognition and action. Kosslyn has argued that this principle has two key correlates. First, coarse-grained visual patterns and minimal sets of fine-grained image features are ordinarily sufficient for basic object identification. Second, once an adequate match has been discovered between a set of image features and the stored record of an object’s shape or function, sensory information not needed for the continued operation of the system in that context is discarded. In this view, visual images are constructed piecemeal, beginning Philosophical Psychology 161 with minimal sets of diagnostic features. Detail is ‘filled in’ only to the extent that it is needed for a given task. For instance, coarse-grained features defining the general shapes of object types are ordinarily sufficient for object identification (e.g., the occlusion boundary and rough configuration of ‘facial features’ in black and white images of Picasso’s ‘Baboon and Young’). However, reaching and grasping may depend on a detailed representation of the structure of an object part. In the former case, there may be no need to attend to fine-grained details (e.g., the features that identify the baboon’s ‘eyes’ as the windshield of a toy car). In the latter case, information about the overall shape of the object may be superfluous to the task and can be discarded once the feature to be grasped has been located. In this regard, perceptual (object recognition), cognitive (object identification and task demands), and motor (reaching and grasping) systems cooperate to determine cue diagnosticity. The net result is an efficient perceptual system directed at the goals of the organism, which naturally filters out irrelevant sensory information. In this context, artists’ spatial schemata represent a novel form of practical knowledge that biases their perception to stimulus features sufficient for adequate artistic production in their chosen medium and style. There are four stages in Kosslyn’s hypothesis testing model for object identification. First, sensory inputs from each fixation are recorded in a short-term ...