Limits...
Wholes and subparts in visual processing of human agency.

Neri P - Proc. Biol. Sci. (2009)

Bottom Line: We measured the perceptual impact of perturbing an agent either by scrambling individual limbs while leaving the relationship between limbs unaffected or conversely by scrambling the relationship between limbs while leaving individual limbs unaffected.Our measurements differed for the two conditions, providing conclusive evidence against a one-stage model.The results were instead consistent with a two-stage processing pathway: an early bottom-up stage where local motion signals are integrated to reconstruct individual limbs (arms and legs), and a subsequent top-down stage where limbs are combined to represent whole agents.

View Article: PubMed Central - PubMed

Affiliation: Institute of Medical Sciences, Aberdeen Medical School, University of Aberdeen, Foresterhill, Aberdeen AB25 2ZD, UK. pn@white.stanford.edu

ABSTRACT
The human visual system is remarkably sensitive to stimuli conveying actions, for example the fighting action between two agents. A central unresolved question is whether each agent is processed as a whole in one stage, or as subparts (e.g. limbs) that are assembled into an agent at a later stage. We measured the perceptual impact of perturbing an agent either by scrambling individual limbs while leaving the relationship between limbs unaffected or conversely by scrambling the relationship between limbs while leaving individual limbs unaffected. Our measurements differed for the two conditions, providing conclusive evidence against a one-stage model. The results were instead consistent with a two-stage processing pathway: an early bottom-up stage where local motion signals are integrated to reconstruct individual limbs (arms and legs), and a subsequent top-down stage where limbs are combined to represent whole agents.

Show MeSH

Related in: MedlinePlus

Limb- and body-selective scrambling of joint trajectories. (a) Schematic of a point-light agent consisting of 12 body joints and the head (the scrambling procedure is illustrated here for one agent only, but the actual stimulus contained two agents). Only three (randomly chosen) limbs were sampled (indicated by black dots) resulting in nine joint trajectories per agent (the head was never sampled). (a,b) In the ‘target’ stimulus, the same temporal segment (indicated by grey shading) was sampled for all trajectories (black rectangles are vertically aligned in (b). (c–f) In the ‘non-target’ stimulus, the nine trajectories were grouped into three triplets (one colour per triplet). The sampling segment for each triplet was shifted by a randomly selected amount (indicated by the coloured arrows) resulting in temporal dephasing (the coloured rectangles are misaligned in (d,f)). No dephasing was present within each triplet (rectangles of the same colour are aligned). (c,d) In the body-scrambling condition, a given triplet comprised joints coming from the same limb, i.e. each triplet corresponded to one limb (c). This triplet assignment resulted in scrambling across limbs (rectangles sampling different limbs are misaligned in (d)), but no scrambling within individual limbs (rectangles sampling the same limb are aligned in (d)). (e,f) In the limb-scrambling condition, a given triplet comprised joints coming from different limbs (e). For example, the red triplet in (e) consists of one joint from the right arm (right shoulder), one joint from the right leg (right foot) and a third joint from the left leg (left knee). This triplet assignment resulted in scrambling within each limb (e.g. misaligned rectangles for the right arm (joints numbered 1–2–3) in (f)), but no scrambling across limbs (the same overall temporal region is sampled for each limb in (f)). The only distinguishing feature between limb scrambling and body scrambling was the way in which the individual joint trajectories were grouped into triplets; all other sampling manipulations were identical. See the electronic supplementary material film for animated versions of the stimuli. (g) Two-stage scheme for a qualitative interpretation of the data. Stage 1 in the processing hierarchy assembles the moving dots into limbs; stage 2 assembles limbs into the full percept of an agent. Limb scrambling (red) places the bottleneck for processing at stage 1; body scrambling (blue) places it at stage 2. (h) Consequently, the non-target stimulus only reaches stage 1 on limb-scrambling trials, but is able to reach stage 2 on body-scrambling trials (filled bars above non-target label). The target stimulus (never scrambled) completes stage 2 (filled bar above target label). The perceptual difference between target and non-target is indicated by the filled double-headed arrows, and is larger on limb-scrambling trials (red) compared with body-scrambling trials (blue). Following inversion, which is hypothesized to knock out stage 2, all stimuli are pushed back to stage 1 in the processing hierarchy (open bars). The perceptual difference between target and non-target is decreased by this manipulation on limb-scrambling trials (compare red filled arrow with open double-headed arrow), but remains unaffected on body-scrambling trials (compare blue filled arrow with open double-headed arrow).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2664361&req=5

fig1: Limb- and body-selective scrambling of joint trajectories. (a) Schematic of a point-light agent consisting of 12 body joints and the head (the scrambling procedure is illustrated here for one agent only, but the actual stimulus contained two agents). Only three (randomly chosen) limbs were sampled (indicated by black dots) resulting in nine joint trajectories per agent (the head was never sampled). (a,b) In the ‘target’ stimulus, the same temporal segment (indicated by grey shading) was sampled for all trajectories (black rectangles are vertically aligned in (b). (c–f) In the ‘non-target’ stimulus, the nine trajectories were grouped into three triplets (one colour per triplet). The sampling segment for each triplet was shifted by a randomly selected amount (indicated by the coloured arrows) resulting in temporal dephasing (the coloured rectangles are misaligned in (d,f)). No dephasing was present within each triplet (rectangles of the same colour are aligned). (c,d) In the body-scrambling condition, a given triplet comprised joints coming from the same limb, i.e. each triplet corresponded to one limb (c). This triplet assignment resulted in scrambling across limbs (rectangles sampling different limbs are misaligned in (d)), but no scrambling within individual limbs (rectangles sampling the same limb are aligned in (d)). (e,f) In the limb-scrambling condition, a given triplet comprised joints coming from different limbs (e). For example, the red triplet in (e) consists of one joint from the right arm (right shoulder), one joint from the right leg (right foot) and a third joint from the left leg (left knee). This triplet assignment resulted in scrambling within each limb (e.g. misaligned rectangles for the right arm (joints numbered 1–2–3) in (f)), but no scrambling across limbs (the same overall temporal region is sampled for each limb in (f)). The only distinguishing feature between limb scrambling and body scrambling was the way in which the individual joint trajectories were grouped into triplets; all other sampling manipulations were identical. See the electronic supplementary material film for animated versions of the stimuli. (g) Two-stage scheme for a qualitative interpretation of the data. Stage 1 in the processing hierarchy assembles the moving dots into limbs; stage 2 assembles limbs into the full percept of an agent. Limb scrambling (red) places the bottleneck for processing at stage 1; body scrambling (blue) places it at stage 2. (h) Consequently, the non-target stimulus only reaches stage 1 on limb-scrambling trials, but is able to reach stage 2 on body-scrambling trials (filled bars above non-target label). The target stimulus (never scrambled) completes stage 2 (filled bar above target label). The perceptual difference between target and non-target is indicated by the filled double-headed arrows, and is larger on limb-scrambling trials (red) compared with body-scrambling trials (blue). Following inversion, which is hypothesized to knock out stage 2, all stimuli are pushed back to stage 1 in the processing hierarchy (open bars). The perceptual difference between target and non-target is decreased by this manipulation on limb-scrambling trials (compare red filled arrow with open double-headed arrow), but remains unaffected on body-scrambling trials (compare blue filled arrow with open double-headed arrow).

Mentions: In the non-target interval, we scrambled individual trajectories by randomly shifting the sampling segment with respect to the originally selected segment (indicated by the grey rectangle in figure 1) within a temporal window of width W centred on the originally selected segment (Neri et al. 2006). We express scrambling strength as (W−S)/S (the same definition used in Bülthoff et al. (1998) for depth scrambling), where S is stimulus duration (width of the grey rectangle). Scrambling could be applied either between (figure 1c,d) or within (figure 1e,f) limbs. We describe the procedure for one agent as both agents were similarly manipulated. The nine joint trajectories (3 limbs×3 joints) corresponding to each agent were grouped into three triplets (indicated by three different colours in figure 1c–f) according to one of two schemes: all joints within a given triplet came from the same limb (body-scrambling condition, figure 1c–d), or all joints within a given triplet came from different limbs (limb-scrambling condition, figure 1e–f). For the latter scheme, which joint from which limb was paired with which joint from a different limb was selected randomly (one specific example is shown in figure 1e,f, but we presented all possible groupings on different trials). After joints were grouped into triplets, joints belonging to the same triplet were scrambled by the same random amount but different triplets were scrambled by different amounts (indicated by the coloured arrows in figure 1d,f).


Wholes and subparts in visual processing of human agency.

Neri P - Proc. Biol. Sci. (2009)

Limb- and body-selective scrambling of joint trajectories. (a) Schematic of a point-light agent consisting of 12 body joints and the head (the scrambling procedure is illustrated here for one agent only, but the actual stimulus contained two agents). Only three (randomly chosen) limbs were sampled (indicated by black dots) resulting in nine joint trajectories per agent (the head was never sampled). (a,b) In the ‘target’ stimulus, the same temporal segment (indicated by grey shading) was sampled for all trajectories (black rectangles are vertically aligned in (b). (c–f) In the ‘non-target’ stimulus, the nine trajectories were grouped into three triplets (one colour per triplet). The sampling segment for each triplet was shifted by a randomly selected amount (indicated by the coloured arrows) resulting in temporal dephasing (the coloured rectangles are misaligned in (d,f)). No dephasing was present within each triplet (rectangles of the same colour are aligned). (c,d) In the body-scrambling condition, a given triplet comprised joints coming from the same limb, i.e. each triplet corresponded to one limb (c). This triplet assignment resulted in scrambling across limbs (rectangles sampling different limbs are misaligned in (d)), but no scrambling within individual limbs (rectangles sampling the same limb are aligned in (d)). (e,f) In the limb-scrambling condition, a given triplet comprised joints coming from different limbs (e). For example, the red triplet in (e) consists of one joint from the right arm (right shoulder), one joint from the right leg (right foot) and a third joint from the left leg (left knee). This triplet assignment resulted in scrambling within each limb (e.g. misaligned rectangles for the right arm (joints numbered 1–2–3) in (f)), but no scrambling across limbs (the same overall temporal region is sampled for each limb in (f)). The only distinguishing feature between limb scrambling and body scrambling was the way in which the individual joint trajectories were grouped into triplets; all other sampling manipulations were identical. See the electronic supplementary material film for animated versions of the stimuli. (g) Two-stage scheme for a qualitative interpretation of the data. Stage 1 in the processing hierarchy assembles the moving dots into limbs; stage 2 assembles limbs into the full percept of an agent. Limb scrambling (red) places the bottleneck for processing at stage 1; body scrambling (blue) places it at stage 2. (h) Consequently, the non-target stimulus only reaches stage 1 on limb-scrambling trials, but is able to reach stage 2 on body-scrambling trials (filled bars above non-target label). The target stimulus (never scrambled) completes stage 2 (filled bar above target label). The perceptual difference between target and non-target is indicated by the filled double-headed arrows, and is larger on limb-scrambling trials (red) compared with body-scrambling trials (blue). Following inversion, which is hypothesized to knock out stage 2, all stimuli are pushed back to stage 1 in the processing hierarchy (open bars). The perceptual difference between target and non-target is decreased by this manipulation on limb-scrambling trials (compare red filled arrow with open double-headed arrow), but remains unaffected on body-scrambling trials (compare blue filled arrow with open double-headed arrow).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2664361&req=5

fig1: Limb- and body-selective scrambling of joint trajectories. (a) Schematic of a point-light agent consisting of 12 body joints and the head (the scrambling procedure is illustrated here for one agent only, but the actual stimulus contained two agents). Only three (randomly chosen) limbs were sampled (indicated by black dots) resulting in nine joint trajectories per agent (the head was never sampled). (a,b) In the ‘target’ stimulus, the same temporal segment (indicated by grey shading) was sampled for all trajectories (black rectangles are vertically aligned in (b). (c–f) In the ‘non-target’ stimulus, the nine trajectories were grouped into three triplets (one colour per triplet). The sampling segment for each triplet was shifted by a randomly selected amount (indicated by the coloured arrows) resulting in temporal dephasing (the coloured rectangles are misaligned in (d,f)). No dephasing was present within each triplet (rectangles of the same colour are aligned). (c,d) In the body-scrambling condition, a given triplet comprised joints coming from the same limb, i.e. each triplet corresponded to one limb (c). This triplet assignment resulted in scrambling across limbs (rectangles sampling different limbs are misaligned in (d)), but no scrambling within individual limbs (rectangles sampling the same limb are aligned in (d)). (e,f) In the limb-scrambling condition, a given triplet comprised joints coming from different limbs (e). For example, the red triplet in (e) consists of one joint from the right arm (right shoulder), one joint from the right leg (right foot) and a third joint from the left leg (left knee). This triplet assignment resulted in scrambling within each limb (e.g. misaligned rectangles for the right arm (joints numbered 1–2–3) in (f)), but no scrambling across limbs (the same overall temporal region is sampled for each limb in (f)). The only distinguishing feature between limb scrambling and body scrambling was the way in which the individual joint trajectories were grouped into triplets; all other sampling manipulations were identical. See the electronic supplementary material film for animated versions of the stimuli. (g) Two-stage scheme for a qualitative interpretation of the data. Stage 1 in the processing hierarchy assembles the moving dots into limbs; stage 2 assembles limbs into the full percept of an agent. Limb scrambling (red) places the bottleneck for processing at stage 1; body scrambling (blue) places it at stage 2. (h) Consequently, the non-target stimulus only reaches stage 1 on limb-scrambling trials, but is able to reach stage 2 on body-scrambling trials (filled bars above non-target label). The target stimulus (never scrambled) completes stage 2 (filled bar above target label). The perceptual difference between target and non-target is indicated by the filled double-headed arrows, and is larger on limb-scrambling trials (red) compared with body-scrambling trials (blue). Following inversion, which is hypothesized to knock out stage 2, all stimuli are pushed back to stage 1 in the processing hierarchy (open bars). The perceptual difference between target and non-target is decreased by this manipulation on limb-scrambling trials (compare red filled arrow with open double-headed arrow), but remains unaffected on body-scrambling trials (compare blue filled arrow with open double-headed arrow).
Mentions: In the non-target interval, we scrambled individual trajectories by randomly shifting the sampling segment with respect to the originally selected segment (indicated by the grey rectangle in figure 1) within a temporal window of width W centred on the originally selected segment (Neri et al. 2006). We express scrambling strength as (W−S)/S (the same definition used in Bülthoff et al. (1998) for depth scrambling), where S is stimulus duration (width of the grey rectangle). Scrambling could be applied either between (figure 1c,d) or within (figure 1e,f) limbs. We describe the procedure for one agent as both agents were similarly manipulated. The nine joint trajectories (3 limbs×3 joints) corresponding to each agent were grouped into three triplets (indicated by three different colours in figure 1c–f) according to one of two schemes: all joints within a given triplet came from the same limb (body-scrambling condition, figure 1c–d), or all joints within a given triplet came from different limbs (limb-scrambling condition, figure 1e–f). For the latter scheme, which joint from which limb was paired with which joint from a different limb was selected randomly (one specific example is shown in figure 1e,f, but we presented all possible groupings on different trials). After joints were grouped into triplets, joints belonging to the same triplet were scrambled by the same random amount but different triplets were scrambled by different amounts (indicated by the coloured arrows in figure 1d,f).

Bottom Line: We measured the perceptual impact of perturbing an agent either by scrambling individual limbs while leaving the relationship between limbs unaffected or conversely by scrambling the relationship between limbs while leaving individual limbs unaffected.Our measurements differed for the two conditions, providing conclusive evidence against a one-stage model.The results were instead consistent with a two-stage processing pathway: an early bottom-up stage where local motion signals are integrated to reconstruct individual limbs (arms and legs), and a subsequent top-down stage where limbs are combined to represent whole agents.

View Article: PubMed Central - PubMed

Affiliation: Institute of Medical Sciences, Aberdeen Medical School, University of Aberdeen, Foresterhill, Aberdeen AB25 2ZD, UK. pn@white.stanford.edu

ABSTRACT
The human visual system is remarkably sensitive to stimuli conveying actions, for example the fighting action between two agents. A central unresolved question is whether each agent is processed as a whole in one stage, or as subparts (e.g. limbs) that are assembled into an agent at a later stage. We measured the perceptual impact of perturbing an agent either by scrambling individual limbs while leaving the relationship between limbs unaffected or conversely by scrambling the relationship between limbs while leaving individual limbs unaffected. Our measurements differed for the two conditions, providing conclusive evidence against a one-stage model. The results were instead consistent with a two-stage processing pathway: an early bottom-up stage where local motion signals are integrated to reconstruct individual limbs (arms and legs), and a subsequent top-down stage where limbs are combined to represent whole agents.

Show MeSH
Related in: MedlinePlus