The Human Visual System

1
The Human Visual System
Vonikakis Vasilios, Antonios Gasteratos
Presented in EUCognition
Democritus University of Thrace 2006
2
The Human Visual System
Optic nerve
Visual Cortex V1, V2
Retina
(ganglion cells)
Biological background
light
3
The eye
?????p??? ?pt??? S?st?µa
4
The photoreceptors

• 3 kinds of cones (long, medium, short) color
  vision (only in bright light photopic vision)
• Rods achromatic vision (in dim light scotopic
  vision)

5
Differences from a ccd

• Only one layer of photoreceptors
• Varying distribution of photoreceptors (Only L
  and M cones in the fovea, only rods in the
  periphery)
• Different ratios of photoreceptors between
  individuals (generally LgtMgtS)
• Hexagonal distribution of photoreceptors
• No refresh rate parallel transmission of visual
  information to the brain

6
What retina sees
Day
Night
7
Basic retinal circuit
photoreceptors

• Ganglion cells are the only output of from the
  retina
• Digital output with an FM modulation (spikes)

Ganglion cell
Output
8
Receptive field

• The number of photoreceptors that a ganglion cell
  sees and the kind of the connection
• Ganglion cells have antagonistic center-surround
  receptive field

9
Center-surround antagonism
10
Center-surround responses
light
No light
light
No light
inhibition
excitation
nothing
11
Center-surround facts

• Ganglion cells are edge detectors they respond
  only to changes and not to uniform areas

• By stimulating only the cells that detect
  differences, the HVS minimizes the number of
  active neurons
• Example Instead of transmitting a sequence of
  long numbers e.g. 2003453, 2003453, 2003455,
  2003451 it transmits only their differences 0,
  0, 2, -2

12
Center-surround advantage

• White paper in dim light reflects less light (is
  darker) than the black letters in bright light
• The absolute value of reflected light is not
  important

• By responding only to differences, ganglion cells
  prevent the white paper from being perceived as
  black

13
Kinds of Ganglion cells
Bcenter - (RG)surround
Blue-Yellow oponency
Gcenter - Rsurround
Rcenter - Gsurround
(RGB)center - (RGB)surround
Achromatic opponency
Photoreceptor mosaic
Biological background
14
Midget ganglion cells
Biological background
15
Midget ganglion cells

• Midget ganglion multiplex 2 signals
• Red-Green chromatic opponency
• Achromatic high acuity (1 cone 1 center of the
  receptive field)

16
Parasol ganglion cells

• Parasol ganglion cells are
• Achromatic
• Have 3 times greater receptive filed
• Respond better to movement

17
Bistratified ganglion cells

• Bistratified ganglion cells
• Carry the Blue Yellow opponency
• Have 3 times greater receptive filed

18
Retinal output

• At least 8 independent and parallel mosaics of
  ganglion cells outputs scan the photoreceptors
  and transmit different information to the visual
  cortex

19
The primary visual cortex V1

• The visual cortex analyses the retinal output in
  3 different and independent maps
• color
• motion-depth
• orientation of edges

20
The primary visual cortex V1

• The visual cortex analyses the retinal output in
  3 different and independent maps
• color
• motion-depth
• orientation of edges

21
Demultiplexing RG in cortex
22
Cell types

• For every position of the visual field there are
  8 different cells that detect chromatic and
  achromatic signals in 2 different scales

23
Cell outputs
Red-Green opponency
original
Blue-Yellow opponency
Achromatic (dark-light)
24
Double opponent cells

• Are formed by combinations of simple
  center-surround cells
• Are excited only by chromatic differences of a
  very specific color (color edges)

25
Responses

• Double opponent cells respond only to very
  specific changes between certain hues (color
  edges)

original
26
Simple Orientation cells

• Elongated receptive fields (formed by
  combinations of center-surround receptive fields)
• 12 different orientations (every 15)
• Detect edges of particular orientations only in a
  very specific position

27
Complex Orientation cells

• Formed by combinations of simple orientation
  cells
• Detect edges of particular orientation anywhere
  in their receptive field

28
Orientation cells

• At every position of the visual field there are
  all possible orientations of an edge
• Every edge excites a particular orientation cell
  in a particular position of the visual cortex

29
Hypercolumns

• For every position of the visual field, all cells
  are grouped into hyper columns
• Every hypercolumn is a complete and independent
  feature detector for a very small part of the
  visual field
• Every hypercolumn contains color cells,
  orientation cells, disparity cells, motion cells

30
Hypercolumns

• Competition exists between cells of the same
  hypercolumn and between hypercolumns

31
Connection of orientation cells

• Orientation cells prefer to be connected with
  others that favor the smooth continuity of
  contours

Biological background
32
Salient contours

• Smooth combinations emerge from the group of
  orientation cells
• This is the first step for contour perception

33
Contour integration

• More complex cells code certain combinations of
  salient orientation cells

34
Feature binding

• All the features (contours, colors, texture,
  depth) are being bind in one perception
• Binding is described by the Gestalt rules e.g.
  common fate rule, proximity rule, similarity rule
  etc.

35
Filling-in the features

• There is a tendency to spatially diffuse strong
  signals over the weak ones
• This way, regions that do not have a strong
  feature get one from a nearby region that has a
  strong one
• Edges act like barriers that stop the diffusions
  of the signals

• There is filling-in for
• Texture
• Color
• Disparity

36
Filling-in illusions
37
Binding to one percept
Object space
binding
38
What Where stream
39
Cell(s) for every object

• Finally there is one cell (or one population of
  cells) that respond only to a very specific
  object
• Every perception of an object (either vision
  triggered or mind triggered) activates these
  cells
• This databank of cells is located at the
  inferior temporal cortex

40
Inferior temporal cortex

• Inferior temporal cortex has columnar
  organization
• Many aspects of an object are stored in
  neighboring columns
• Similar objects are stored in neighboring rows

41
Inferior temporal cortex

• Every object is stored in the object space in
  many rotated versions
• but we are trained only to the versions we
  usually see

42
Attention models
43
Attention models
44
Thank you!