How the "Rule of Thirds" is Wrong: Let us Count the ways
Effects of object facing direction and implied motion on preferences for spatial composition
Palmer, Gardner, and Wickens (2008) found an 'inward bias' in aesthetic preferences for the position of a single object inside a frame that depended on its facing direction: right-facing objects were preferred left of center and left-facing objects were preferred right of center. They hypothesized that participants would also prefer objects that characteristically move forward (e.g., people, dogs, and cars) to be located farther from the center of the frame than objects that are characteristically static (e.g., flowers, chairs, and teapots) to provide more space for their forward motion, but no difference was detected between these two object classes. In the present experiments, we tested whether the motion direction and speed of self-propelled moving objects influenced the preferred horizontal position of forward facing objects. We used images of left- and right-facing humans, horses, and cars that depicted different motion directions and speeds. The results showed that motion direction trumped facing direction for forward/backward divers, in that divers facing one direction and moving in the opposite direction (i.e., backward divers) were preferred to be facing out of the frame but moving into it. Additional conditions investigated the effects of implied speed using images of standing, walking, jogging, and running humans and standing, walking, trotting, and galloping horses. Inward biases were again present, and decreased as implied speed increased. These results suggest that in addition to facing direction, both implied motion speed and motion direction affect preferred horizontal positioning.
Representational Fit in Position and Perspective: A Unified Aesthetic Account
Previous research on aesthetic preference for spatial compositions has shown robust and systematic preferences for object locations within frames, such as the center bias, the inward bias, and various ecological biases (Palmer, Gardner, & Wickens, 2008; Gardner & Palmer, VSS-2006, VSS-2008, VSS-2009). These preferences can be dramatically altered, however, by changing contextual meaning through different titles for the same picture (Gardner & Palmer, VSS-2009). Perspective is a similar factor: People also prefer canonical perspectives (Palmer, Rosch, & Chase, 1981) when rating their aesthetic response to pictures of everyday objects (Khalil & McBeath, VSS-2006), but these preferences can also be shifted by changing the context through different titles. Our theoretical account of preference for the composition that best fits the context - which we call "representational fit" (Gardner & Palmer, VSS-2009) - can explain not only preferences in the "default" case, where the goal is simply to present the focal object(s) optimally in a way that best captures its most salient features (e.g., as in stock photography, see Gardner, Fowlkes, Nothelfer, and Palmer, VSS-2008), but also more nuanced and realistic cases in which there is a meaning associated with the image beyond its explicit image content. The current research examines several aspects of this preference for representational fit. People prefer non-standard compositions (with regard to position and/or perspective) more than standard compositions, so long as there is a context that justifies the unexpected composition. Put another way, there is greater artistic value in novelty and violating expectations, provided that the results are meaningful and coherent. These results provide strong evidence for representational fit as an aesthetic theory that unifies fluency accounts, where the default context prevails (Reber, Schwarz, & Winkielman, 2004), with classic aesthetic accounts in terms of novelty and violating expectations, where a nonstandard meaning is intended or inferred.
Aesthetics of Spatial Composition: Semantic Effects in Two-Object Pictures
Previous research on aesthetic response to spatial composition of simple pictures examined preferences for the horizontal position of single objects within a rectangular frame (Palmer, Gardner & Wickens, 2007). The results revealed a center bias for front-facing objects and an inward bias for left or right-facing objects. The current studies examined aesthetic preferences for compositions containing two objects. Each picture contained one stationary object, whose position was fixed, and one movable object, whose position was adjusted by the participant to create the most aesthetically pleasing composition. The stationary object was presented at one of five equally-spaced locations along a horizontal axis. In the first experiment, four vertically symmetrical objects without a facing direction - two short, wide objects (sponge, cake) and two tall thin objects (plastic bottle of liquid dish soap, bottle of sparkling wine) - were presented in pairs consisting of one short, wide object and one tall, thin object. The center points of the preferred position for the movable objects were then binned to compute frequency histograms of their preferred positions. When the two objects were related (wine and cake or liquid soap and sponge), people generally placed the movable object close to the fixed object, whereas when they were unrelated (wine and sponge or liquid soap and cake), people generally placed the movable object far away from the fixed object. In a second study the same data were collected in a between-participants design - such that each participant saw only a single pair of objects - to control for possible demand characteristics arising from the same participant seeing both related and unrelated pairs of objects. The second experiment also assessed the effects of semantic relatedness on the preferred direction of facing.
Aesthetic Preferences in the Size of Images of Real-world Objects
In previous research, Konkle and Oliva (VSS-2009) found that the preferred visual size ("canonical size") of a picture of an object is proportional to the log of its known physical size: Small physical objects are preferred when their images are small within a frame and large physical objects are preferred when their images are large within a frame. They employed withinparticipant designs using multiple objects in several different tasks, including a perceptual preference task in which they asked participants to adjust the image size so that the object "looks best." Because of concerns about how the instructions were interpreted (is the image that "looks best" the one at which it "looks most like itself" or the one that is "most aesthetically pleasing") and possible demand characteristics (the same person seeing multiple objects of different sizes may implicitly feel pressured to make their relative sizes consistent), we studied image size effects on aesthetic judgments using a two-alternative forced-choice method in both within- and between-participant designs, asking them to choose the picture that you "like best." In Experiment 1, participants saw all possible pairs of images depicting the same object at six different sizes for twelve real-world objects that varied in physical size. Consistent with Konkle and Oliva's findings, participants preferred small objects to be smaller in the frame and large objects to be larger, regardless of whether they saw only a single object (the between-participant design) or all objects intermixed (the within-participant design). In Experiment 2, we examined whether this effect would still be evident if the amount of visual detail present at different sizes was equated by "posterizing" the images. Here the ecological bias toward relative size effects disappeared. Our findings indicate that multiple factors interact in determining aesthetic responses to images of different sizes.
Representational Transparency in Aesthetic Judgments of Spatial Composition:
Effects of Object Position and Size
Previous research has shown robust, systematic aesthetic preferences for the horizontal position and facing direction of single objects within rectangular frames (Palmer, Gardner & Wickens, 2008; Gardner & Palmer, VSS-2006, VSS-2008). People prefer an object to be laterally positioned near the center (the "center bias") and to face into, rather than out of, the frame (the "inward bias"). Similar, but more complex, biases occur in the vertical dimension: a "lower bias" for objects supported from below and viewed from above (a bowl on a table), an "upper bias" for objects supported from above and viewed from below (a light fixture on a ceiling), and a "center bias" for symmetrical images of gravitationally unsupported objects (a flying eagle viewed from directly below or above). The object's characteristic ground-relative position in the world also affects people's preferences for vertical placement: eagles are preferred higher and stingrays lower in the frame. Real-world compatibility in the size domain also affects aesthetic judgments: a mouse picture is preferred when it is smaller within the frame and an elephant when it is larger (Konkle & Oliva VSS-2007). Canonical perspectives (Palmer, Rosch & Chase, 1981) also produce higher preference ratings for pictures of objects (Khalil & McBeath, VSS-2006). These effects can be unified by the "representational transparency" hypothesis: observers prefer images in which the spatial characteristics of depicted objects in the world are optimally reflected in analogous spatial properties of the image. This has both a real-world-position component (eagles higher, stingrays lower) and a viewer-relative component (objects viewed from below are preferred higher, objects viewed from above are preferred lower). Representational transparency provides a reasonable first-order approximation of default expectations for people's aesthetic responses, but greater aesthetic value often requires violating these expectations in meaningful ways that reflect the intentions of the artist.
Exploring aesthetic principles of spatial composition through stock photography
Past research in our laboratory (Palmer, Gardner & Wickens, in press; Palmer & Gardner, VSS 2007) has shown robust and systematic aesthetic preferences for the horizontal position and direction of a single object within a frame. In particular, people prefer the object to be laterally positioned near the center of the frame (the "center bias") and to face into, rather than out of, the frame (the "inward bias"). In the present research we extend these findings with experimentally manipulated images to the vertical dimension, where we find a strong "lower bias" for objects supported from below (e.g., a cup or bowl) and an "upper bias" for those supported from above (e.g., a ceiling light). We also investigated the extent to which these horizontal and vertical biases are manifest in aesthetically pleasing natural images outside the laboratory by analyzing images from the Corel database of stock photography. Observers viewed hundreds of images that they judged to contain just one or two focal objects and indicated where they perceived the center of the visible portions of these objects to be located. Using these data, we examined evidence for the center, inward, lower, and upper biases found in our previous laboratory research separately for one- and two-object pictures. We also tested models of people's judgments about the location of the center of the visible portion of an object (e.g. bounding-box, center of mass, geometrical center, etc.).
Joint Effects of Height-in-the-Picture-Plane and Distance-Relative-to-the-Horizon in Pictorial Depth Perception
Height in the picture plane (or height in the field) is known as a powerful and salient depth cue (Dunn, 1965). Rock (1975) found that the distance to the horizon also affects judgments of depth: objects that are located closer to the horizon line are seen as more distant. Height in the picture plane can work independently of distance to the horizon: the horizon itself can be placed at different heights in the frame, affecting the depth of objects, even if distance to the horizon is held constant. Additionally, distance to the horizon can vary if the horizon line changes but the object position in the frame stays the same. However, when the object is located on the ceiling plane, these two depth cues can be seen to conflict - an object on the ceiling plane may be high in the picture plane but far from the horizon. Using a 2AFC paradigm, we examined the depth cues of height in the picture plane and distance to the horizon to determine the way in which information from these depth cues is obtained and combined.
Framing Aesthetic Judgments
Painters, photographers, and graphic designers regularly face the problem of how to frame the subjects of their creations in aesthetically pleasing ways. We investigated people's aesthetic responses to the position, facing direction, and size of single objects within rectangular frames using free choices in taking actual photographs and 2AFC preferences. The experiments tested the validity of rules of thumb taught in the visual arts, most of which have never been tested experimentally. One example is the "facing rule:" if the subject of the work has horizontal directionality (e.g., a sideview of a person, car, or teapot), it should point into rather than out of the frame. An experiment testing this rule examined subjects' aesthetic preferences for pictures of objects pointing into and out of the frame as a function of their position and directionality. In the directional (sideview) conditions, preferences were found for objects pointing into versus out of the frame. In both the directional and nondirectional (frontview) conditions, subjects tended to prefer objects positioned at the center of the frame. Further experiments examined preferences for the size of objects relative to the frame and its interaction with position and directionality. The results are discussed in terms of the power of the center in visual art (Arnheim, 1988). People prefer the subject to be located in the center of the frame, but if an object is not in the center, they prefer it to be oriented toward the center.
All of the above research is funded in part by NSF Grant BCS-0745820 to Stephen E. Palmer and by a gift from Google, Inc.