1 Autostereograms and Autostereoscopic Perception Leslie Blaha, Christopher Honey, and Eric Nichols October 31, 2005 When the Magic Eye phenomenon swept through American culture in the 1990s, we all marveled at our ability to make three-dimensional objects “magically” pop out of seemingly nonsensical two-dimensional pictures. From one perspective it should not be surprising for people to see three-dimensional objects; we rely on that ability to navigate the world on a daily basis. We also regularly use multiple image cues, including shading, shadows, occlusion, color, and relative size, to infer three-dimensionality from two-dimensional depictions of real-world scenes (Julesz, 1962). Yet, we do not experience those images as if they were three-dimensional objects in the way the Magic Eye images appear to be three-dimensional; nor do typical pictures produce oscillating two- and three-dimensional percepts. The Magic Eye experience clearly represents a unique visual phenomenon which occurs with the full, conscious awareness of the observer. Magic Eye images are better known as autostereograms, random-dot stereogram images that produce three-dimensional percepts without the use of a stereoscope. This is distinct from both stereoptic vision of natural three-dimensional objects and stereoscopic vision of stereograms presented in a stereoscope (Bishop, 1996). We shall refer to such perception as autostereoscopic. What is an autostereogram? In traditional stereoscopic perception, stereograms can be viewed as three-dimensional images by providing two side-by-side views of a three-dimensional scene, rendered from slightly different viewpoints. Stereoscopes facilitate focusing of the eyes depths other than the surface of a two-dimensional stereogram image, allowing the two separate images to overlap in the center of the field of vision (Kimmel, 2002). Once the correct depth is achieved, it may seem to the viewer as if the two images are different due to the horizontal retinal disparities. Then the two images can fuse into a coherent three-dimensional percept where horizontal disparities in the images provide important depth cues. Some individuals can view regular stereograms (Figure 2A) by adjusting eye convergence to the correct depth without the aid of a stereoscope. Autostereograms obviate the use of a stereoscope by amalgamating two stereogram patterns into a single image, incorporating either a random texture (Figure 2B) or a repeating two-dimensional pattern (Figure 2C). They have the additional experimental advantage of looking rather like random noise until the correct convergence and fusion of a three-dimensional image is achieved.2 Figure 1. Depth perception via distances between repeating patterns. Diagram reproduced from Kimmel (2002, reproduced without permission). a.3 b. c. Figure 2 Stereograms and autostereograms. (a) Stereogram image of Manhattan, c. 1909 – 1910. Each side would be presented to a visual field in a stereoscopic experiment. (b) Autostereogram employing a random-dot pattern. Image contains an embedded M.C. Escher figure. (c) Two autostereograms employing repeated two-dimensional patterns. The first image depicts an embedded whirlpool; the second is a wave pattern. Autostereoscopic perception, then, is thought to involve two distinct phases: synthesis and maintenance of the three-dimensional percept. The synthesis stage can be thought of as the foveal fusion of the appropriate points within the image; this foveation4 focuses vision at the proper depth to create the three-dimensional percept. Previous studies have referred to this as the three-dimensional search phase (Burgess, Rehman, & Williams, 2003), onset (Portas, et al., 2000), or fusion/“the construction of a unified cyclopean representation” (Revonsuo, et al., 1997, p. 3870). Once the three-dimensional percept has been formed, the maintenance phase is the span of time in which an observer is aware of that percept (i.e. is aware of the object popping out of the background). Observers can generally control this phase by maintaining foveation on the object. This phase is also referred to as the three-dimensional phase (Burgess, Rehman, & Williams, 2003) or the “constant experience of a stable gestalt” (Revonsuo, et al., 1997, p. 3870). It is our goal to determine if distinct neural mechanisms support this differentiation of the synthesis and maintenance processes in autostereoscopic perceptions, examining not only which areas are involved in both phases, but also the extent of temporal and spatial dissociation of these neural mechanisms. Advantages of Autostereograms In studying phenomenological awareness of visual perception, it is desirable to employ a framework in which we can use minimal physical stimulus manipulations to produce maximally different perceptual results. Kim and Blake (In press) review seven different experimental paradigms allowing experimenters to manipulate the visual awareness of stimuli. As part of their analysis they enumerate 8 criteria to be considered in evaluating the effectiveness of techniques that erase a stimulus from awareness.1 Their first 4 criteria relate to the generality of the stimulus used in the technique: the variety of stimuli that can be erased, the size of stimuli that can be erased, the size of the area in the visual field in which stimuli can be erased and the lack of temporal constraint on stimuli. The second set of 4 criteria relate to the effectiveness of the technique in erasing awareness: whether the absence/presence of the stimulus is unambiguous, whether the erasure of the perceived stimulus can be achieved for an invariant physical stimulus, whether the duration of absence/presence state is long enough for effective brain-image capture, whether the absence/presence of the stimulus is temporally predictable. Table 1 presents an excerpt from a table of Kim and Blake (In press), summarizing their assessment of the various paradigms along each of the 8 dimensions. We have added a row summarizing the properties of autostereoscopic perception for the purpose of easy comparison. Note that an autostereogram can be understood as an extreme version of a bistable image. Table 1 illustrates that the two techniques are rate similarly on most criteria. The key practical differences are that autostereograms can be used for a greater variety of stimulus structures and can be sustained for a longer period of time; on the