8
1 Motion Perception Issues in Visual Cognition George Mather What visual processes mediate motion perception? An influential traditional theory of motion perception views apparent motion as a perceptual inference. “…an intelligent solution to the problem posed by the successive appearance and disappearance of spatially separated objects.” (Sigman & Rock, 1974) Wagon wheel aliasing: the wheels appear to rotate backwards when their displacement exceeds the sampling rate of the movie display. Can be explained by nearest-neighbour matching. The wagon wheel illusion Neural motion sensors In the early 1960’s, Hubel and Wiesel found cells in V1 that were tuned to motion direction. Responses are velocity dependent, not displacement dependent. Stimulus Cell Response Receptive fields The receptive field of direction-selective neurons can be described as elongated in space-time (plot is based on recordings from a simple cell in cat V1; McLean & Palmer, 1989). t x + - - + - - T ime Space Evidence for motion detectors in human vision Discontinuity effects Adaptation

What visual processes mediate - University of Sussex

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

1

Motion Perception

Issues in Visual CognitionGeorge Mather

What visual processes mediate motion perception?

♦ An influential traditional theory of motion perception views apparent motion as a perceptual inference.

♦ “…an intelligent solution to the problem posed by the successive appearance and disappearance of spatially separated objects.” (Sigman & Rock, 1974)

♦ Wagon wheel aliasing: the wheels appear to rotate backwards when their displacement exceeds the sampling rate of the movie display.

♦ Can be explained by nearest-neighbour matching.

����

����

����

The wagon wheel illusion Neural motion sensors

♦ In the early 1960’s, Hubel and Wiesel found cells in V1 that were tuned to motion direction.

♦ Responses are velocity dependent, not displacement dependent.

Stimulus Cell Response

Receptive fields

♦ The receptive field of direction-selective neurons can be described as elongated in space-time (plot is based on recordings from a simple cell in cat V1; McLean & Palmer, 1989).

∆∆t

x

+ --

+ --

T ime

Space

Evidence for motion detectors in human vision

♦ Discontinuity effects♦ Adaptation

2

Random Block Kinematograms

♦ RBKs are a standard psychophysical tool for studying early motion detecting processes.

Frame 1

Frame 2

Random Block Kinematograms

♦ Motion discrimination is limited to:♦ small spatial

displacements

♦ Short ISIs (not shown)

Frame 1

Frame 2

40

50

60

70

80

90

100

10 20 30 40 50 60

Displacement (min arc)

Me

an P

erc

en

t C

orr

ect

Cells vs. perception

♦ Newsome et al. (1986) found that cortical direction-selective cells respond to apparent or discontinuous movement (within limits) – filled symbols.

♦ The limits of perception (open circles) are similar to the limits of cell response.

10

100

0 .1 1

Th reshold

V1

MT

TemporalInterval(msec)

S patia l Inte rva l (deg)

M OT ION

N O MOT ION

Motion Discontinuity

T im e

S pace∆∆S ∆∆S

∆∆T∆∆T

♦ Beyond a certain limit, discontinuous motion (sometimes called apparent or stroboscopic motion) is ineffective as a stimulus for motion detectors.

Motion after-effect

♦ Adaptation to motion in one direction results in the perception of opposite apparent motion in subsequently viewed stationary patterns.

Motion after-effect

♦ Fixate on the branch in the middle of the Falls

3

Explanation of the MAE

♦ The MAE can be explained by opponent-processing or a tug-of-war between motion sensors tuned to different directions.

♦ The MAE has become a standard technique for investigating the properties of motion sensors (Mather et al., 1998).

Discontinuity reduces MAE duration

♦ Several studies have shown that MAE duration declines as the discontinuity of the adapting motion increases (data from Anstisand Mather, 1985)

0.1

1

10

0 5 10 15

Mean MAE Duration (sec)

Separa tion (m in arc)

0

50

100

150

200

250

300

350

0 .1 1

S pa tial In te rval (deg)

Ap paren t M ot ion

0-5% Erro rs

In the early 1970’s, Braddick found that results with RDKs were very different from those with classical displays.

Two-processes? The two-process model

♦ Short-range process – Responsible for motion detection over short distances and time intervals. Identified with neural motion detectors.

♦ Long-range process – Responsible for motion detection over long distances and time intervals. Identified with perceptual inferences and/or attentional tracking.

♦ The effect reverses when the fundamental is removed.

����

����

The wagon-wheel effect can be explained by motion detector responses Wagon wheel aliasing

♦ Inspection of the stimulus profile indicates that the effect is inconsistent with feature tracking.

F1

F2F1

F2

4

Wagon wheel aliasing

♦ Different components shift in different directions.

♦ The largest component in a missing fundamental pattern moves backwards.

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0 0.1 0.2 0.3 0.4 0.5

Fundamental Phase Shift (cycles)

Ph

ase

Sh

ift

(cyc

les

f

3f

5f

7f

9f

-1.25

-0.75

-0.25

0.25

0.75

1.25

Position

Am

plit

ude

F1

F3

F5

F7

F9

SUM

-0.75

-0.5

-0.25

0

0.25

0.5

0.75

Position

Am

plit

ude

F1

F3

F5

F7

F9

SUM

Other evidence for a position-based process?

♦ Colour-defined bars only weakly stimulate motion detectors, but can be tracked attentively (Cavanagh, 1992).

Other evidence for a position-based process?

♦ Apparent direction in random patterns depends on stimulus onset asynchrony (Boulton and Baker, 1993).

Cortical Processing

♦ There are 30 cortical areas responsive to visual stimuli

♦ This figure is from Felleman and van Essen (1991).

Cortical Processing

♦ There are about 300 projections between areas.

♦ Difficult to infer the processing hierarchy.

♦ A pattern emerges for the largest seven areas.

♦ Dorsal or parietal route (V3/MT/MST)

♦ Ventral or temporal route (V4/IT)

Two processing streams

♦ Dorsal or parietal route (V3/MT/MST)

♦ Ventral or temporal route (V4/IT)

5

Two processing streams

♦ Several researchers have proposed that the dorsal and ventral streams have different processing roles.

Conscious visionUnconscious action control

WhatWhere

FormMotion

VENTRAL/TEMPORALDORSAL/PARIETAL

Evidence for two processing streams

♦ Several lines of evidence support a difference in function. But there is no agreement on the three variants (review in Bruce at al., 2003).

Visual form agnosia with intact action

Blindsight

Lesions in V4 affect colour and pattern vision, and attention

Lesions in MT affect motion perception but not pattern

vision

V4 has few motion detectors but is rich in colour cells

MT and MST are rich in motion detectors

VENTRALDORSAL

Phenomenology of Biological Motion

♦ Biological motion displays (Johansson, 1973) represent walking humans using isolated points of light located on major joints.

Biological Motion

♦ Biological motion can depict many actions very effectively

Biological Motion

♦ Biological motion can depict many actions very effectively (Mather and West, 1993).

Biological Motion

6

Gender discrimination in biological motion

♦ Gender can be judged from biological motion

How is biological motion processed? Two possibilities…

♦ Position-based♦ Tracking the position of each point

to infer body structure.♦ Computer vision systems have

used this scheme.

Biological Motion: Position

Natural fit for the ventral stream/long-range system

Shoulder

Elbow

W rist

Hip

Knee

Ankle

1 2 3 4Shoulder

Elbow

W rist

Hip

Knee

Ankle

1 2 3 4

How is biological motion processed? Two possibilities…

♦ Velocity-based♦ The movement of

each point activates low-level motion detectors.

♦ Recognition is based on learned patterns of movement.

♦ Natural fit for the dorsal stream/short-range system.

Shoulder

Elbow

W rist

Hip

Knee

Ankle

1 4

Effect of discontinuity on biological motion

♦ Discrimination of coherent vs. incoherent walkers (circles), and walker direction (squares) is sensitive to inter-stimulus interval.

♦ Data from Mather et al. (1992) 50

55

60

65

70

75

80

85

90

-20 0 20 40 60 80 100 120 140

C oherence

Direction

M ean Pe rcen t Co rrect

Inte r-stimu lus Inte rval (msec)

Gender discrimination in biological motion

♦ An influential explanation is based on the structural difference between male and female torsos.

♦ But is dynamic information important?

♦ In my lab we measured gender discrimination in brief views of biological motion displays

7

Gender discrimination in biological motion

♦ Discrimination of gender is best in brief views containing higher velocity movements.

♦ Again consistent with responses in motion sensors.

♦ Data from Mather and Murdoch (1994) 0

0 .5

1

1 .5

2

2 .5

3

0 45 90 135 180 225 270 315 360

M ean

Velocity D if fe rence

Sensitivity (d')

Starting P hase (deg)

Biological motion processing: physiological correlates

♦ Dorsal route includes MT, MST, and STS

Single-cell recordings

♦ Cells in STS respond to locomotion.

♦ Data from Perrett et al. (1990) show responses from a cell in Macaque temporal cortex to views of biological motion.

♦ Response is selective for forward-facing, forward movement.

Brain scans

♦ Human brain scans find activity along STS while viewing biological motion.

♦ Consistent with activation of motion processing areas.

♦ Figure is from Puce and Perrett (2003).

Is the issue of biological motion processing resolved?

♦ No –– Some researchers still claim high-level processes

involving cognition and/or attention play a role.

– Others claim that the ventral stream (form analysis) plays a role.

Effect of trial duration

♦ Thornton et al. (1998) repeated Mather et al.’s (1992) experiment on ISI, using longer trial durations.

♦ Results showed performance remained high.

8

Effect of mask♦ Thornton et al. (1998)

used a scrambled walker mask at long durations.

♦ Again, performance remained high.

♦ They conclude that attentional tracking is involved in biological motion processing (‘long-range’ processing).

Where to go from here?

♦ We shall explore some other papers in the seminar…