Spatial vision John Greenwood Department of Experimental Psychology NEUR3045 Contact: john.greenwood@ucl.ac.uk 1
Today • What is spatial vision? • Physiology of spatial vision • The dimensions of spatial vision: Fourier analysis • Orientation • Adaptation and population coding • Spatial frequency • The contrast sensitivity function (CSF) • Population coding for spatial frequency • Foveal vs peripheral vision • Acuity and crowding 2
What is spatial vision? • Our perception of the spatial V1 distribution of light across the visual field • The building blocks of object perception • The early stages of visual LGN processing Retina 3
Spatial vision: LGN • Retinal ganglion cells and neurons in On-centre the Lateral Geniculate Nucleus have centre-surround receptive fields - - + • Both on-centre and off-centre subtypes • Can highlight regions of change • (i.e. transitions from light to dark or Off-centre dark to light) • But this does not give selectivity to - + + the orientation of edges 4
Spatial vision: V1 • Hubel & Wiesel (1962) found - + + - + - - orientation selectivity in the + - - + + - + - - + + primary visual cortex (V1) - + - - LGN + + • Cells respond to particular V1 orientations of edges & lines • Have a preferred orientation that produces maximal spike/firing rate (Schiller et al., 1976) • Likely built from particular combinations of centre-surround LGN neurons (Hubel & Wiesel, 1968) 5
The dimensions of spatial vision • We can consider the ‘building blocks of spatial vision’ via Fourier analysis • Fourier (1822) showed that any signal can be decomposed into a sum of sine waves at different frequencies, amplitudes and phases 6
Sine wave amplitude • Amplitude for a sine wave grating gives luminance contrast • The difference between light and dark regions in the scene 100% 75% 50% 25% 0% 100 50 0 7
Sine wave phase • Phase determines the point at which variations occur in space, e.g. the starting point of the cycle • Represented in radians with a cyclical structure • Determines the position of edges in the scene π /2 3 π /2 2 π π 0 8
Sine wave orientation • For two-dimensional images we also need to consider the orientation of the sine wave 0° 45° 90° 135° 180° • Orientation is certainly a key dimension for visual processing and we’ll return to this shortly 9
Sine wave spatial frequency • Spatial frequency determines the variations across space • Reported as the number of cycles in a spatial region (peak to peak) • Captures the fine vs. coarse detail in an image 1 cyc/image 2 cyc/image 4 cyc/image 8 cyc/image 16 cyc/image Low High 10
Summing the components • How do we make an image using sine waves? • Sum all of the component sine waves together • Easiest example: a square wave • How do you get a square wave from sine wave components? 11
From to .. Fundamental (F) 3F 5F 7F • Take a sine wave Components with matched spatial frequency: the fundamental • Add the odd harmonics Fundamental (F) F+3F+5F+7F (increasing SF) F+3F F+3F+5F with decreasing amplitude Sum 12
Summing the ideas • Fourier analysis tells us that we can break the visual scene down into component wave forms characterised by: • Amplitude (contrast) • Phase (position) • Spatial frequency (size) • Orientation (orientation…) • How are these dimensions encoded in the visual system? • Let’s look at two aspects: • Orientation • Spatial frequency 13
Filtering an image with filters similar to the receptive fields of V1 cells gives us orientation energy at a range of spatial frequencies Spatial frequency (low to high) Orientation ∑ Original ∑ 14
Local orientation • How do we encode orientation across the visual field? • One way to examine this: adaptation • Gibson (1937): prolonged viewing of one orientation reduces sensitivity to that orientation, and produces repulsion in the perceived tilt of dissimilar gratings Adapt Test 15
Orientation adaptation • Adaptation reduces sensitivity to the adapting orientation • e.g. higher contrast required for detection • Can be attributed to reduced sensitivity of the underlying neurons s u s Pre-adapt Post-adapt u l u l u m m i t i S t S 16
Local orientation • Gibson (1937): prolonged viewing of one orientation reduces sensitivity to that orientation, and produces repulsion in the perceived tilt of dissimilar gratings • i.e. adaptation reduces sensitivity to the adapting orientation (performance) • It can also alter the perceived orientation (appearance), which we call the tilt aftereffect Adapt Test 17
Tilt aftereffect • Subsequent dissimilar orientations appear repulsed away • Produces a shift in the peak response away from the adaptor • Suggests population coding of orientation (Blakemore et al., 1971) Pre-adapt Post-adapt t r t s o s e e t T T p a d A shifted peak 18
Two key principles • Adaptation • Reduces neural responses to continued stimulation and enhances responses to novel stimuli (redundancy reduction) • Population coding • Adapting to one orientation influences the perception of others • Our perception of orientation is inferred from the population of neural responses, e.g. as the peak • Allows a resolution higher than the sensitivity of individual neurons 19
Orientation in context • Suppressive effects are also seen outside the local region • Contrast surround effects especially apparent with matched orientation & spatial frequency (Chubb et al., 1989) 20
Surround suppression • Can again be attributed to reduced sensitivity of the underlying neurons, via connections from adjacent neurons s u s u l u l u m m i t i S t S 21
Context effects: Tilt contrast • With dissimilar orientations can also see a shift in perceived orientation - similar to the effects of adaptation and the tilt aftereffect (Gibson, 1937) 22
Tilt contrast • Adjacent orientations appear repulsed from one another • Also accounted for by shifts in population response, induced by adjacent neurons (Blakemore et al., 1970) d n u t t s o s e e r T T r u S 23
Surround effects • Likely mediated by intracortical connections in V1 • Around 90% or neurons in V1 are suppressed by the activity of their neighbours (Jones et al , 2001) • Suppression minimises the response to homogeneous regions and highlights differences • Another instance of redundancy reduction (across space rather than time) • Minimises metabolic costs of neural firing (Laughlin et al., 1998) 24
Spatial frequency • Fourier analysis also gives us a way to think about scale • Images contain information at different spatial frequencies • Which of these components is visible to an observer? • With Fourier analysis we can take a broader view of the image content that is visible to a given observer 1 cyc/image 2 cyc/image 4 cyc/image 8 cyc/image 16 cyc/image 25
SF in natural scenes • What does spatial frequency mean for natural scenes? • Low-pass filtering: allow only the lowest SFs to be visible (broad blobby things) • High-pass filtering: allow only the highest SFs to be visible (edges & fine detail) 26
A note on units • How do we characterise these spatial variations? • Cycles/image is OK for theoretical Fourier analyses • But for visual perception, image size on the retina is affected by both size and distance • Need to measure retinal size • Calculated as degrees of visual angle, where tan( α ) = Height/Distance • For SF gives cycles/degree α 27
Contrast sensitivity functions • Campbell & Robson (1968): high • Measured contrast sensitivity at a range of spatial frequencies Highest • Contrast sensitivity function visible SF (CSF) peaks around 4 c/deg (acuity cutoff) • Sensitivity is not greatest for 1 deg. uniform regions (low SF)! • Sensitivity also drops for high SFs - highest visible spatial low frequency is our acuity cutoff • Altogether defines our ‘window of visibility’ 28
A depiction of the CSF • We visualise low the CSF by plotting Contrast (amplitude) contrast against spatial frequency • Note: peak in the middle & the drop in visibility on high either side low high Spatial frequency 29
What produces the CSF? • Why do we show this pattern of sensitivity? • Campbell & Robson (1968) hypothesised that the visual system is composed of spatial frequency channels - each sensitive to a restricted range of SFs • Blakemore & Campbell (1969) tested this using adaptation Adapt Test or Test 30
CSF adaptation: predictions • Adaptation reduces sensitivity to contrast • But does it affect all SFs or just those of the adaptor? Single Channel Multiple Channels Adapting SF Adapting SF 31
Multiple SF channels • Adaptation to a sine grating with 7.1 cycles per degree • Sensitivity is strongly reduced at the adapted SF and nearby values Adapting SF • No effect for SF values at the extremes of the range • Consistent with multiple channels for spatial frequency • Evidence that we separate the visual scene into its Fourier components (at least for SF) 32
Recommend
More recommend
