W3 Spatial Vision: Lines and Texture

New Concepts

• area V1 of visual cortex

• detecting oriented lines and edges (w simple + complex cells)

• inhibition over space (via hyper-complex cells)

• inhibition over space in terms of:

  • texture segmentation

  • simultaneous tilt illusion

  • simultaneous size illusion

• inhibition over times in terms of:

  • tilt after-effect

  • size after-effect

Oriented edges and bars: Cortex area V1

• Classical view - Hubel and Wiesel (1950s)#

  • won Nobel prize

  • cat - 1962, monkey - 1968

• Simple cells = orientation sensitive, off + on regions

• Complex cells = orientation sensitive, prefers thin lines, no off region

• Hypercomplex = same as complex cells, but prefer short lines

Cortex area V1: Simple Cells

• simple cells have receptive fields

• structured on / off regions

• different orientations and sizes

• because of these diff orientations = allows V1 cells to detect stimuli at diff orientations + sized

Orientation selectivity in simple cells

• whichever angle you present light to a cell in retina/LGN = always excites the same art of the receptive field

• in simple cell (V1) = the cell will fire maximally when the stimulus excited only the ‘on’ part of the receptive field

  • highest red dot is when light is vertical through on section

Position sensitivity in simple cells

• V1 simple cells can detect the position of an object in space

• because as the light moves into the ‘on’ part of the cell = excites it maximally

Size sensitivity (spatial frequency)

• width + spatial frequency

• wide object = bc it overlaps on/off regions = cancels out activity

• as stimulus becomes more narrow / spatial frequency increases = overlaps with on part of receptive field = fires maximally

• as it becomes more narrow = part of stimulus overlaps on/off regions = decreases firing of the cell

Making a simple cell

• simple cells constructed by combining the output of many concentric receptive fields from the LGN

• imagine the ‘on’ sections of a concentric receptive field sending their output to the on/excitatory part of the V1 simple cell

• and all the inhibitory parts of the concentric LGN cells = sending output to inhibitory part of the V1 simple cell

Summary: Simple cells

• Orientation selective

• some position selective

• some size selective

• Separate on / off regions

• Length summation = strong response to long bar / little response to short bar

= SO, simple cells in V1 respond to edges at particular locations + orientations within the visual field

Cortex area V1: complex cells

• they are more complex cells in V1

• high resting output (no stimulus = more activity)

• receptive fields were defined but unstructured = no on/off receptive field

• also orientation sensitive

• also spatial frequency sensitive

• not sensitive to line position within the recepetive field

• more for abstract objects

Making a complex cell

• classic view = complex cells are connected to many simple cells by an OR functions

• a stimulus is detected by any of these simple cells = makes the complex cells fire

• complex cells represent a more abstract type of visual info - partially independent of location within the visual field

Complex cells

• orientation selective

• some size selective

• NO separate on / off regions

• not position selective

• length summation

Hyper-complex cells

• AKA end-stopping cells (because they can detect the length of a stimulus/object)

• they are like complex cells

• but they are sensitive to line length + orientation

• cells turned off by stimulus outside its ‘classical’ receptive field

• receptive field is organised like = the r field has two inhibitory cells either side = can turn the cell off when stimulus gets too long (end-stopping)

Hyper-complex cells: how might inhibition work?

made of 3 complex cells

Length sensitivity

Texture segmentation: what is texture?

• Texture = spatial detail at a finer scale than that which the observer currently defines as the object scale (finer detail than the object itself)

• Textures are defined by statistical properties rather than absolute values

• if 2 textures have the same statistical properties (as far as a human is concerned) = they are the same

• If 2 textures have different stat. properties = they are different + will be separable

• This enables use to see edges of objects by changes in their surface markings

• Great for breaking camouflage

Texture Segmentation: Texton Theory WRONG

• Texton = texture atom

• Pattern elements will looks different (be segmentable) if they have different numbers of textons

• Broken Ls and Ts have 6 textons (2 lines, 4 line endings)

• Crosses have 7 textons (2 lines, 4 endings, 1 crossing)

• THIS IS WRONG bc segmentation based on line crossings and line endings is easily disrupted

Texture Segmentations: Nothdurft (1990/91)

• Texture segmentation based on line crossings + endings - is more affected by visual manipulations than by orientation

• E.g. random variations in size, position disrupts segmentation based on crossings + endings (but not segmentation based on orientation)

So, orientation differences and size (local contrast) may be the only real ‘textons’

Texture segmentation: Orientation contrast model (Nothdurft, 1990)

They all have 2 line endings but due to local contrast we can segment them

Segmentation by size contrast

How does the brain detect orientation?

• Centre surround cells

• there are 2 classes of cells in V1 which are sensitive to orientation:

  • single opponent cell

  • double opponent cell

Centre surround cells for orientation

Single opponent cell:

• can respond to a single orientation

• can also respond when that orientation contrasts with a different one

Double opponent cell:

• can respond to more than one orientation (vertical + horizontal)

• also when stimuli contrast with the orientation of objects surrounding it

Single + double opponent cell

• 3rd bar = centre and surround are contrasting

• 5th bar = only double cell responds bc its horizontal stimulus

Simultaneous tilt illusion

• bc of centre surround cells

• tile-illusion works like simultaneous contrast but with orientation replacing luminance

• the lines in periphery make the lines in centre seem tiled when they are actually straight

• the receptive field for detecting orientation is creating spatial inhibition

Does simultaneous contrast work for size?

• Ebbinghaus illusion

• peripheral part of stimulus turns of cell less = makes centre look larger

• (left) peripheral circles are larger = turn off receptive field for size more = makes circle in the centre look smaller

Inhibition over time: tilt after-effects

• same as when you see opponent colour after looking at colour for a long time

• inhibition over time in area V1 = ‘tile after-effect’

• some cells inhibited after long time = the opposite orientation cells are more active

• this tricks you into thinking a vertical stimulus is tilted in the opposite direction

Tilt after-effect

• perceived orientation is determined by a group of cells sensitive to different orientations = the neural code

• prolonged viewing (adapting) of one orientation for a long time = inhibition

• the response of the active cells is decreased by adaptation / inhibition

• the inhibition takes time to take effect + lasts after the stimulus goes away

• the new orientation shown is not seen properly (you think its shifted to the opposite direction of the original / adapter stimulus)

Summary: Key Concepts

• Cells in area V1 that respond to different orientations:

  • simple cells = sensitive to orientation + position

  • complex cells = sensitive to orientation only

  • hyper-complex cells = sensitive to orientation + length

• hyper-complex cells may lead to perception of illusory contours in area V2

• texture segmentation due to orientation + size contrast

• centre-surround cells for orientation in V1 = lead to simultaneous tilt-illusion

• similar mechanisms for size = lead to Ebbinghaus illusion

• inhibition over time among orientation cells = lead to tile after-effect + size after effect