Stereopsis in animals: evolution, function and mechanisms by Nityananda V, Read JCA, NityanandaRead2017.pdf (0.8 MiB) - Stereopsis is the computation of depth information from views
acquired simultaneously from different points in space. For many
years, stereopsis was thought to be confined to primates and other
mammals with front-facing eyes. However, stereopsis has now been
demonstrated in many other animals, including lateral-eyed prey
mammals, birds, amphibians and invertebrates. The diversity of
animals known to have stereo vision allows us to begin to investigate
ideas about its evolution and the underlying selective pressures in
different animals. It also further prompts the question of whether all
animals have evolved essentially the same algorithms to implement
stereopsis. If so, this must be the best way to do stereo vision, and
should be implemented by engineers in machine stereopsis.
Conversely, if animals have evolved a range of stereo algorithms in
response to different pressures, that could inspire novel forms of
machine stereopsis appropriate for distinct environments, tasks or
constraints. As a first step towards addressing these ideas, we here
review our current knowledge of stereo vision in animals, with a view
towards outlining common principles about the evolution, function
and mechanisms of stereo vision across the animal kingdom. We
conclude by outlining avenues for future work, including research into
possible new mechanisms of stereo vision, with implications for
machine vision and the role of stereopsis in the evolution of
camouflage.
Blog Archives
Visual Perception: Neural Networks for Stereopsis
Visual Perception: Neural Networks for Stereopsis by Read JCA, Cumming BG, ReadCumming2017.pdf (0.4 MiB) - This is a comment article on Welchman and Goncalves (2017): ‘‘What not’’ detectors help the brain see in depth. Curr. Biol. 27, 1403–1412.
True stereoscopic 3D cannot be simulated by shifting 2D content off the screen plane
True stereoscopic 3D cannot be simulated by shifting 2D content off the screen plane by Hands P, Read JCA, HandsRead2017.pdf (0.9 MiB) - money by including brief sections of 2D content displayed with a uniform disparity, i.e. the 2D image is
geometrically shifted behind the screen plane. This manipulation is believed to produce an illusion of
depth which, while not as powerful as true S3D, is nevertheless more compelling than simple 2D. Our
study examined whether this belief is correct. 30 s clips from a nature documentary were shown in
the original S3D, in ordinary 2D and in shifted versions of S3D and 2D. Participants were asked to determine
the impression of depth on a 7 point Likert scale. There was a clear and highly significant difference
between the S3D depth perception (mean 6.03) and the shifted 2D depth perception (mean 4.13)
(P = 0.002, ANOVA). There was no difference between ordinary 2D presented on the screen plane, and
the shifted 2D. We conclude that the shifted 2D method not only fails to mimic the depth effect of true
S3D, it in fact has no benefit over ordinary 2D in terms of the depth illusion created. This could impact
viewing habits of people who notice the difference in depth quality.
geometrically shifted behind the screen plane. This manipulation is believed to produce an illusion of
depth which, while not as powerful as true S3D, is nevertheless more compelling than simple 2D. Our
study examined whether this belief is correct. 30 s clips from a nature documentary were shown in
the original S3D, in ordinary 2D and in shifted versions of S3D and 2D. Participants were asked to determine
the impression of depth on a 7 point Likert scale. There was a clear and highly significant difference
between the S3D depth perception (mean 6.03) and the shifted 2D depth perception (mean 4.13)
(P = 0.002, ANOVA). There was no difference between ordinary 2D presented on the screen plane, and
the shifted 2D. We conclude that the shifted 2D method not only fails to mimic the depth effect of true
S3D, it in fact has no benefit over ordinary 2D in terms of the depth illusion created. This could impact
viewing habits of people who notice the difference in depth quality.
Overestimation of stereo thresholds by the TNO stereotest is not due to global stereopsis.
Overestimation of stereo thresholds by the TNO stereotest is not due to global stereopsis. by Vancleef K, Read JCA, Herbert W, Goodship N, Woodhouse M, Serrano-Pedraza I, VancleefReadHerbertGoodshipWoodhouseSerranoPedraza2017_2.pdf (18 KiB) - Purpose
It has been repeatedly shown that the TNO stereotest overestimates stereo threshold compared to other clinical stereotests. In the current study, we test whether this overestimation can be attributed to a distinction between ‘global’ (or ‘cyclopean’) and ‘local’ (feature or contour-based) stereopsis.
Methods
We compared stereo thresholds of a global (TNO) and a local clinical stereotest (Randot Circles). In addition, a global and a local psychophysical stereotest were added to the design. One hundred and forty-nine children between 4 and 16 years old were included in the study.
Results
Stereo threshold estimates with TNO were a factor of two higher than with any of the other stereotests. No significant differences were found between the other tests. Bland-Altman analyses also indicated low agreement between TNO and the other stereotests, especially for higher stereo threshold estimates. Simulations indicated that the TNO test protocol and test disparities can account for part of this effect.
Discussion
The results indicate that the global – local distinction is an unlikely explanation for the overestimated thresholds of TNO. Test protocol and disparities are one contributing factor. Potential additional factors include the nature of the task (TNO requires depth discrimination rather than detection) and the use of anaglyph red/green 3D glasses rather than polarizing filters, which may reduce binocular fusion.
It has been repeatedly shown that the TNO stereotest overestimates stereo threshold compared to other clinical stereotests. In the current study, we test whether this overestimation can be attributed to a distinction between ‘global’ (or ‘cyclopean’) and ‘local’ (feature or contour-based) stereopsis.
Methods
We compared stereo thresholds of a global (TNO) and a local clinical stereotest (Randot Circles). In addition, a global and a local psychophysical stereotest were added to the design. One hundred and forty-nine children between 4 and 16 years old were included in the study.
Results
Stereo threshold estimates with TNO were a factor of two higher than with any of the other stereotests. No significant differences were found between the other tests. Bland-Altman analyses also indicated low agreement between TNO and the other stereotests, especially for higher stereo threshold estimates. Simulations indicated that the TNO test protocol and test disparities can account for part of this effect.
Discussion
The results indicate that the global – local distinction is an unlikely explanation for the overestimated thresholds of TNO. Test protocol and disparities are one contributing factor. Potential additional factors include the nature of the task (TNO requires depth discrimination rather than detection) and the use of anaglyph red/green 3D glasses rather than polarizing filters, which may reduce binocular fusion.
Avoiding monocular artifacts in clinical stereotests presented on column-interleaved digital stereoscopic displays
Avoiding monocular artifacts in clinical stereotests presented on column-interleaved digital stereoscopic displays by Serrano-Pedraza I, Vancleef K, Read JCA, SerranoPedrazaVancleefRead.pdf (1.5 MiB) - New forms of stereoscopic 3-D technology offer vision
scientists new opportunities for research, but also
come with distinct problems. Here we consider
autostereo displays where the two eyes’ images are
spatially interleaved in alternating columns of pixels
and no glasses or special optics are required. Columninterleaved
displays produce an excellent stereoscopic
effect, but subtle changes in the angle of view can
increase cross talk or even interchange the left and
right eyes’ images. This creates several challenges to
the presentation of cyclopean stereograms (containing
structure which is only detectable by binocular vision).
We discuss the potential artifacts, including one that is
unique to column-interleaved displays, whereby scene
elements such as dots in a random-dot stereogram
appear wider or narrower depending on the sign of
their disparity. We derive an algorithm for creating
stimuli which are free from this artifact.We show that
this and other artifacts can be avoided by (a) using a
task which is robust to disparity-sign inversion—for
example, a disparity-detection rather than
discrimination task—(b) using our proposed algorithm
to ensure that parallax is applied symmetrically on the
column-interleaved display, and (c) using a dynamic
stimulus to avoid monocular artifacts from motion
parallax. In order to test our recommendations, we
performed two experiments using a stereoacuity task
implemented with a parallax-barrier tablet. Our
results confirm that these recommendations eliminate
the artifacts. We believe that these recommendations
will be useful to vision scientists interested in running
stereo psychophysics experiments using parallaxbarrier
and other column-interleaved digital displays
scientists new opportunities for research, but also
come with distinct problems. Here we consider
autostereo displays where the two eyes’ images are
spatially interleaved in alternating columns of pixels
and no glasses or special optics are required. Columninterleaved
displays produce an excellent stereoscopic
effect, but subtle changes in the angle of view can
increase cross talk or even interchange the left and
right eyes’ images. This creates several challenges to
the presentation of cyclopean stereograms (containing
structure which is only detectable by binocular vision).
We discuss the potential artifacts, including one that is
unique to column-interleaved displays, whereby scene
elements such as dots in a random-dot stereogram
appear wider or narrower depending on the sign of
their disparity. We derive an algorithm for creating
stimuli which are free from this artifact.We show that
this and other artifacts can be avoided by (a) using a
task which is robust to disparity-sign inversion—for
example, a disparity-detection rather than
discrimination task—(b) using our proposed algorithm
to ensure that parallax is applied symmetrically on the
column-interleaved display, and (c) using a dynamic
stimulus to avoid monocular artifacts from motion
parallax. In order to test our recommendations, we
performed two experiments using a stereoacuity task
implemented with a parallax-barrier tablet. Our
results confirm that these recommendations eliminate
the artifacts. We believe that these recommendations
will be useful to vision scientists interested in running
stereo psychophysics experiments using parallaxbarrier
and other column-interleaved digital displays
Neurons in Striate Cortex Signal Disparity in Half-Matched Random-Dot Stereograms
Neurons in Striate Cortex Signal Disparity in Half-Matched Random-Dot Stereograms by Henriksen S, Read JCA, Cumming BG, HenriksenReadCumming2016.PDF (0.8 MiB) - Human stereopsis can operate in dense “cyclopean” images containing no monocular objects. This is believed to depend on the computation of binocular correlation by neurons in primary visual cortex (V1). The observation that humans perceive depth in half-matched random-dot stereograms, although these stimuli have no net correlation, has led to the proposition that human depth perception in these stimuli depends on a distinct “matching” computation possibly performed in extrastriate cortex. However, recording from disparity-selective neurons in V1 of fixating monkeys, we found that they are in fact able to signal disparity in half-matched stimuli. We present a simple model that explains these results. This reinstates the view that disparity-selective neurons in V1 provide the initial substrate for perception in dense cyclopean stimuli, and strongly suggests that separate correlation and matching computations are not necessary to explain existing data on mixed correlation stereograms.
Visual Perception: A Novel Difference Channel in Binocular Vision
Visual Perception: A Novel Difference Channel in Binocular Vision by Henriksen S, Read JCA, HenriksenRead2016.pdf (0.9 MiB) - A "Dispatch", i.e. a comment article, about May & Zhaoping 2016 "Efficient Coding Theory Predicts a Tilt Aftereffect from Viewing Untilted Patterns". Our summary: "A recent study provides compelling evidence that binocular vision uses two separate channels; one channel adds the images from the two eyes, and the other subtracts them. "
A single mechanism can account for human perception of depth in mixed correlation random dot stereograms
A single mechanism can account for human perception of depth in mixed correlation random dot stereograms by Henriksen S, Cumming BG, Read JCA, HenriksenCummingRead2016.PDF (0.7 MiB) - Relating neural activity to perception is one of the most challenging tasks in neuroscience.
Stereopsis—the ability of many animals to see in stereoscopic 3D—is a particularly tractable
problem because the computational and geometric challenges faced by the brain are
very well understood. In essence, the brain has to work out which elements in the left eye’s
image correspond to which in the right image. This process is believed to begin in primary
visual cortex (V1). It has long been believed that neurons in V1 achieve this by computing
the correlation between small patches of each eye’s image. However, recent psychophysical
experiments have reported depth perception in stimuli for which this correlation is zero,
suggesting that another mechanism might be responsible for matching the left and right
images in this case. In this article, we show how a simple modification to model neurons
that compute correlation can account for depth perception in these stimuli. Our model
cells mimic the response properties of real cells in the primate brain, and importantly, we
show that a perceptual decision model that uses these cells as its basic elements can capture
the performance of human observers on a series of visual tasks. That is, our computer
model of a brain area, based on experimental data about real neurons and using only a single
type of depth computation, successfully explains and predicts human depth judgments
in novel stimuli. This reconciles the properties of human depth perception with the properties
of neurons in V1, bringing us closer to understanding how neuronal activity causes
perception.
Stereopsis—the ability of many animals to see in stereoscopic 3D—is a particularly tractable
problem because the computational and geometric challenges faced by the brain are
very well understood. In essence, the brain has to work out which elements in the left eye’s
image correspond to which in the right image. This process is believed to begin in primary
visual cortex (V1). It has long been believed that neurons in V1 achieve this by computing
the correlation between small patches of each eye’s image. However, recent psychophysical
experiments have reported depth perception in stimuli for which this correlation is zero,
suggesting that another mechanism might be responsible for matching the left and right
images in this case. In this article, we show how a simple modification to model neurons
that compute correlation can account for depth perception in these stimuli. Our model
cells mimic the response properties of real cells in the primate brain, and importantly, we
show that a perceptual decision model that uses these cells as its basic elements can capture
the performance of human observers on a series of visual tasks. That is, our computer
model of a brain area, based on experimental data about real neurons and using only a single
type of depth computation, successfully explains and predicts human depth judgments
in novel stimuli. This reconciles the properties of human depth perception with the properties
of neurons in V1, bringing us closer to understanding how neuronal activity causes
perception.
The stereoscopic anisotropy develops during childhood.
The stereoscopic anisotropy develops during childhood. by Serrano-Pedraza I, Herbert W, Villa-Laso L, Widdall M, Vancleef K, Read JCA, SerranoPedrazaHerbertVillaLasoWiddallVancleefRead2016.pdf (1.5 MiB) - PURPOSE:
Human vision has a puzzling stereoscopic anisotropy: horizontal depth corrugations are easier to detect than vertical depth corrugations. To date, little is known about the function or the underlying mechanism responsible for this anisotropy. Here, we aim to find out whether this anisotropy is independent of age. To answer this, we compare detection thresholds for horizontal and vertical depth corrugations as a function of age.
METHODS:
The depth corrugations were defined solely by the horizontal disparity of random dot patterns. The disparities depicted a horizontal or vertical sinusoidal depth corrugation of spatial frequency 0.1 cyc/deg. Detection thresholds were obtained using Bayesian adaptive staircases from a total of 159 subjects aged from 3 to 73 years. For each participant we computed the anisotropy index, defined as the log10-ratio of the detection threshold for vertical corrugations divided by that for horizontal.
RESULTS:
Anisotropy index was highly variable between individuals but was positive in 87% of the participants. There was a significant correlation between anisotropy index and log-age (r = 0.21, P = 0.008) mainly driven by a significant difference between children and adults. In 67 children aged 3 to 13 years, the mean anisotropy index was 0.34 ± 0.38 (mean ± SD, meaning that vertical thresholds were on average 2.2 times the horizontal ones), compared with 0.59 ± 0.55 in 84 adults aged 18 to 73 years (vertical 3.9 times horizontal). This was mainly driven by a decline in the sensitivity to vertical corrugations. Children had poorer stereoacuity than adults, but had similar sensitivity to adults for horizontal corrugations and were actually more sensitive than adults to vertical corrugations.
CONCLUSIONS:
The fact that adults show stronger stereo anisotropy than children raises the possibility that visual experience plays a critical role in developing and strengthening the stereo anisotropy.
Human vision has a puzzling stereoscopic anisotropy: horizontal depth corrugations are easier to detect than vertical depth corrugations. To date, little is known about the function or the underlying mechanism responsible for this anisotropy. Here, we aim to find out whether this anisotropy is independent of age. To answer this, we compare detection thresholds for horizontal and vertical depth corrugations as a function of age.
METHODS:
The depth corrugations were defined solely by the horizontal disparity of random dot patterns. The disparities depicted a horizontal or vertical sinusoidal depth corrugation of spatial frequency 0.1 cyc/deg. Detection thresholds were obtained using Bayesian adaptive staircases from a total of 159 subjects aged from 3 to 73 years. For each participant we computed the anisotropy index, defined as the log10-ratio of the detection threshold for vertical corrugations divided by that for horizontal.
RESULTS:
Anisotropy index was highly variable between individuals but was positive in 87% of the participants. There was a significant correlation between anisotropy index and log-age (r = 0.21, P = 0.008) mainly driven by a significant difference between children and adults. In 67 children aged 3 to 13 years, the mean anisotropy index was 0.34 ± 0.38 (mean ± SD, meaning that vertical thresholds were on average 2.2 times the horizontal ones), compared with 0.59 ± 0.55 in 84 adults aged 18 to 73 years (vertical 3.9 times horizontal). This was mainly driven by a decline in the sensitivity to vertical corrugations. Children had poorer stereoacuity than adults, but had similar sensitivity to adults for horizontal corrugations and were actually more sensitive than adults to vertical corrugations.
CONCLUSIONS:
The fact that adults show stronger stereo anisotropy than children raises the possibility that visual experience plays a critical role in developing and strengthening the stereo anisotropy.
Small or far away? Size and distance perception in the praying mantis.
Small or far away? Size and distance perception in the praying mantis. by Nityananda V, Bissianna G, Tarawneh G, Read JCA, Nityananda_et_al2015_PhilTrans_PostReview.pdf (1.5 MiB) - Stereo or '3D' vision is an important but costly process seen in several evolutionarily distinct lineages including primates, birds and insects. Many selective advantages could have led to the evolution of stereo vision, including range finding, camouflage breaking and estimation of object size. In this paper, we investigate the possibility that stereo vision enables praying mantises to estimate the size of prey by using a combination of disparity cues and angular size cues. We used a recently developed insect 3D cinema paradigm to present mantises with virtual prey having differing disparity and angular size cues. We predicted that if they were able to use these cues to gauge the absolute size of objects, we should see evidence for size constancy where they would strike preferentially at prey of a particular physical size, across a range of simulated distances. We found that mantises struck most often when disparity cues implied a prey distance of 2.5 cm; increasing the implied distance caused a significant reduction in the number of strikes. We, however, found no evidence for size constancy. There was a significant interaction effect of the simulated distance and angular size on the number of strikes made by the mantis but this was not in the direction predicted by size constancy. This indicates that mantises do not use their stereo vision to estimate object size. We conclude that other selective advantages, not size constancy, have driven the evolution of stereo vision in the praying mantis.This article is part of the themed issue 'Vision in our three-dimensional world'.