Posted on August 11, 2011

Crosstalk reduces the amount of depth seen in 3D images of natural scenes


Inna Tsirlin, Robert S. Allison, Laurie M. Wilcox

Crosstalk remains an important determinant of S3D image quality. Defined as the leakage of one eye’s image into the image of the other eye crosstalk affects all commercially available stereoscopic viewing systems. It is well established that crosstalk decreases perceived image quality and causes image distortion (Seuntiens et al. 2005, Wilcox & Stewart, 2002). Moreover, visual comfort decreases and perceived workload increases with increasing crosstalk (Kooi and Toet, 2004; Lambooij, 2010; Pala et al. 2007).  In a series of experiments we have shown that crosstalk also affects perceived depth magnitude (Tsirlin et al. 2011a; Tsirlin et al. 2011b). In our previous experiments we used two white bars on a black background, and measured the perceived depth between the bars as a function of disparity and degree of crosstalk.  The data showed that as crosstalk increased perceived depth decreased. This effect was intensified for larger disparities. We found the effect was present regardless of whether the ghost image was spatially separated from, or overlapped with, the original image. The experiments described here extend our previous work to complex images of natural scenes.

We controlled crosstalk levels by simulating them in images presented on a zero-crosstalk mirror stereoscope display. The stimulus was a color image of our laboratory that showed a cluttered scene composed of furniture and objects. The observers were asked to estimate the amount of stereoscopic depth between pairs of objects in the scene. We used two different estimation methods – a virtual measurement scale and a disparity probe. Data show that, as was the case with simple line stimuli, depth in this natural scene was dramatically affected by crosstalk. As crosstalk increased perceived depth decreased; an effect that grew with increasing disparity. Interestingly, observers overestimated the depth in displays that contained no crosstalk. We propose that this overestimation is the result of the presence of pictorial cues to depth (perspective, texture gradients etc.) and familiarity with the real size of the objects depicted in the image. This hypothesis was confirmed by a control experiment where observers estimated depth in the same natural scene presented in 2D instead of S3D. Although there was no stereoscopic depth in this case, observers still perceived some depth between object pairs. Some observers spontaneously reported nausea and headaches after performing the task in S3D, which confirms previous findings that crosstalk causes discomfort in viewers (Kooi and Toet, 2004).  Taken together these results show that our previous findings generalize to natural scenes showing that crosstalk affects perceived depth magnitude even in the presence of pictorial depth cues.  Our data underscore the fact that crosstalk is a serious challenge to the quality of S3D media and has to be carefully addressed by display manufacturers.

Posted on August 11, 2011

Motion in depth constancy in stereoscopic displays

Submitted to SDA 2012 (San Francisco, January 2012)

Sidrah Laldin, Laurie Wilcox, Carly Hylton, Rob Allison

In stereoscopic vision, there is non-linear mapping between real space and disparity. In a stereoscopic 3D scene, this non-linear mapping could produce distortions of space when camera geometry differs from natural stereoscopic geometry. When the viewing distance and zero screen parallax setting are held constant and interaxial separation (IA) is varied, there is an asymmetric distortion in the mapping of stereoscopic to real space. If an object traverses this space at constant velocity, one might anticipate distortion of the perceived trajectory. This prediction is based on the premise that when the object traverses compressed space, it should appear to move slower than when it passes through expanded space. In addition, this effect should depend on the saliency of the depth information in the scene.  To determine if the predicted distortions are in fact perceived, we assessed observers’ percepts of acceleration and deceleration using an animation of a ball moving in depth through a simulated environment, viewed stereoscopically.

The method of limits was used to measure transition points between perceived acceleration and deceleration as a function of IA and context (textured vs. non-textured background).Eleven observers with normal binocular vision were tested using four IAs (35, 57.4, 65.7, and 68.21mm). The range of acceleration / deceleration rates presented was selected to bracket the predicted values based on the IA and the viewing geometry. Two environments were used to provide different levels of monocular depth cues, specifically an untextured and a tiled ground plane. For each environment and IA combination, four measures were made of the transition points between perceived acceleration and deceleration. For two of these measures, the series of clips began with an obviously accelerating object and progressed to an obviously decelerating object. The participants’ task was to identify the point at which the percept changed from accelerating to decelerating. In the other two measures, the converse procedure was used to identify the deceleration to acceleration transition.

Based on binocular geometry, we predicted that the transition points would shift toward deceleration for small IA and towards acceleration for large IA. This effect should be modulated by monocular depth cues. However, the average transition values were not influenced by IA or the simulated environment. These data suggest that observers are able to discount distortions of stereoscopic space in interpreting the trajectory of objects moving through simple environments. It remains to be seen if velocity constancy will be similarly maintained in more complex scenes or scenes containing multiple moving objects. These results have important implications for the rendering or capture of effective stereoscopic 3D content.

Posted on August 11, 2011

34th European Conference on Visual Perception, Toulouse France

“The effect of interocular separation on perceived depth from disparity in complex scenes”

K Benzeroual, S R. Laldin, L M. Wilcox, R S. Allison

The geometry of stereopsis makes straightforward predictions regarding the effect of increasing an observer’s simulated interocular distance (IO) on perceived depth. Our aim is to characterize the effect of IO on perceived depth, and its dependence on scene complexity and screen size. In Experiment 1 we used S3D movies of an indoor scene, shot with three camera separations (0.25″, 1″ and 1.7″). We displayed this footage on two screens (54″ and 22″) maintaining a constant visual angle. A reference scene with an IO of 1″ was displayed for 5s followed by the test scene. Participants (n=10) were asked to estimate the distances between four pairs of objects in the scene relative to the reference. Contrary to expectations, there was no consistent effect of IO, and all participants perceived more depth on the smaller screen. In Experiment 2 we used static line stimuli, with no real-world context. The same set of conditions was evaluated; all observers now perceived more depth in the larger display and there was a clear dependence on IO. The presence of multiple realistic depth cues has significant and complex effects on perceived depth from binocular disparity; effects that are not obvious from binocular geometry.