Vision Flashcards
% of Cortex associated with Vision AND why is vision hard?
• 30% of the cortex is dedicated to vision alone; 46 identified visual areas located throughout the brain
Why is vision hard?
o Curse of dimensionality – challenge of decoding the brain; too many dimensions to explore
o Inverse problem – why pattern-matching doesn’t work
o Empirical theory of perception – visual system is optimized to each animals environment
Eye Anatomy
– 6 muscles aid in movement (4 rectus; 2 oblique)
o Sclera – protective white outer covering of the eye
o Cornea – clear part of the sclera
o Pupil – where light enters the eye
o Iris – colored part of eye that controls the diameter and thus how much light enters the eye
o Optic Disk – located in the retina and is where the optic nerve leaves the eye
NO photoreceptors and thus the location of our blind spot
o Retina – located in posterior wall of eye
Macula densa – highly pigmented area
• Fovea – point of the eye with highest density of photoreceptors and where we have the highest visual acuity; no blood vessels
o Foveola – ONLY cones; in center of fovea
Basic Ambiguity
– input to the eye is always changing based on illumination, reflectance, occlusion, distance (size), object rotation (view angle)
o Reflectance depends on: angle and material properties; illumination diffusivity and pattern; transparency (fog)
Simultaneous Contrast & White’s Effect
– light entering the eye due to the white background inhibits light coming from darker gray dot, making the gray dot seem lighter as well
Curse of Dimensionality and Inverse Problem and Empirical Theory of Perception
Curse of Dimensionality – impossible to characterize a neurons receptive field (what stimulus makes the neuron fire) because there are too many different possibilities to explore
Inverse Problem – for any single retinal image, there are infinite number of sources that could’ve produced that image; we never see the same image twice
Empirical Theory of Perception – visual system of all animals is optimized for environment we live in
o More sensitive to horizontal lines than diagonal lines
o Higher acuity in lower visual field than the upper visual field
Light Path and Accomodation
cornea pupil lens inner chamber of eye back of retina
o Light is refracted as it passes through the cornea and lens
o Lens thickness affects the amount of refraction to allow the image to focus on the back of retina
o Accommodation – governed by ciliary muscles; shape of lens is a ball naturally
Objects far - ciliary muscles relax; zonule fibers contract lens flattens to allow for less refraction
Objects close – ciliary muscles contract; zonule fibers relax lens curves to allow for more refraction
Cells of Retina
o Photoreceptors – share common structure with an inner and outer segment but different functions
o Outer segment – contains membranous disks w/ photopigment; where transduction occurs
o Inner segment – contains the cell nucleus which connects to the bipolar, horizontal cells
o Rods – specialized for low light vision; less spatial acuity; not color sensitive
o Cones – less sensitive to light; high spatial resolution; fast adaptation; color sensitive
Scotopic vs. Mesopic vs. Photopic Vision
Scotopic – when luminance is very low (darkness to starlight)
No color vision and poor acuity because you are below cone threshold
Mesopic – from starlight to moonlight/indoor lighting
Cone threshold is reached therefore both rods and cones are responding
Photopic – from just before indoor lighting and anything brighter
Rod saturation; allowing for highest resolution with color vision and best acuity
Highest levels of luminance can cause photoreceptor damage
Light Perception
o Rods peak within the green wavelength at 496nm
o 3 Cones – blue/purple range (S) (419nm), green/yellow (M) (531nm), & red/orange (L) (559nm)
o Specific cone photoreceptors respond to photons of different wavelengths allowing our brain to see different colors
Clinical Correlation: Genetics and Photoreceptors
o Blue pigment gene is on an autosomal chromosome
o Genes that code for red and green pigment cones are on the X chromosome
Easy for crossover to occur or gene loss due to close proximity; results in color blindness
Males have one X chromosome; therefore are more prone to red-green color blindness
Light vs. Dark at Molecular Level including Adaptation
o Depolarize to darkness and hyperpolarize to light
o Dark Current – high cGMP levels inside the rod cell; cGMP binds to Na+ channels allowing them to stay open and depolarize; also allows some Ca+ to enter
Light Adaptation: Ca+ flows in through Na+/Ca+ gated channels and inhibits guanylyl cyclase (the enzyme that synthesizes cGMP from GTP) therefore reducing cGMP eyes adapt to lower levels of light
o Light – photons hit rhodopsin within rod cells activate Gprotein cascade activates cGMP phosphodiesterase cGMP levels decrease blockage of Na+ channels hyperpolarization
Retina Neurons: Rods and Cones
o Cones – use bipolar cells
Off-center bipolar cell – send signal to directly ganglion cell in the same format it received; contain ionotropic glutamate receptor
On-center bipolar cell – negative conversion (inverts the signal) and then sends signal to ganglion cell; contain metabotropic glutamate receptor
o Rods – use amacrine cells
Retina Neurons: Horizontal, Amacrine, and Ganlion Cells
o Horizontal cells – lateral inhibition with respect to contrast; uses GABA
o Amacrine cells -lateral inhibition w/ respect to contrast; uses GABA/glycine; 70% input to RGCs
o Ganglion cells – FIRST/ONLY spiking cells in the retina – convert the signal from analog (graded response) to a digital (spiking response)
Most anterior layer of retina and responsible for sending signal to the optic nerve
• 100 million photoreceptors but only 1 million fibers in the optic nerve
Don’t respond to light; respond to contrast and edges
Retinal Ganglion Cells
o Midget (parvocellular/P pathway) - small cell body & area; higher acuity- A cells; 80% of RGC
Concerned with shape, size, and color
o Parasol (magnocellular/M pathway) – large cell body & area; poor acuity- B cells; 10% of RGC
Concerned with motion and light
o Bistratified (Koniocellular/K pathway) – 10% of RGC; concerned with blue light
o Photosensitive (to daylight) ganglion cells projecting to superchiasmatic nucleus (SCN) & LGN
o Other RGC projecting to superior colliculus
Ganglion Cells and Vision
o Center-surround Organization – starts at bipolar cells and horizontal cells
On-center ganglion cell responds to stimuli in center; inhibited in periphery
Off-center ganglion cell responds to stimuli on periphery; inhibited in center
Recodes/condenses the visual information to fit through the optic nerve
o Able to process info b/z vision is redundant & predictable w/ temporal & spatial correlations
Pixels that are close in space and time are very similar (like HD picture on TV)
Ganglion cells use this by de-emphasizing the redundancy (areas of constant illumination) and emphasize areas of edges or areas with contrasts
Retinal-Ganglion Cell Projections
o Optic nerve optic chiasm split and cross the midline so that each eye projects to both sides of the cortex based on R/L visual fields
Hypothalamus – circadian rhythms via the suprachiasmatic nucleus
Pretectum – pupillary light reflex
Superior colliculus – eye movements; “blindsight”
• Blindsight – phenomenon in which blind patients with V1 lesions still maintain the reflexes to respond to visual stimuli
Retinogeniculostriate Pathway
o Signal is transmitted through optic nerve to the lateral geniculate nucleus (LGN) of the thalamus and then radiates out to primary visual cortex (V1)
o LGN – primary center for visual input; signals segregated by eye & cell type; 6 different layers
• Sensitive to contralateral eye vs. ipsilateral eye
2 - Magnocellular pathway (large cell body) – concerned with movement, low light
4 - Parvocellular pathway (small cell body) – concerned with shape, size, and color
• Concerned with red/ green light
o 2 pathways leave the LGN – each responds to different properties and is optimized for different aspects of vision
Clinical Correlation: Blindness in R eye Bitemporal Hemianopsia L. homonymous Hemianopsia Upper Quadrant Hemianopsia Homonymous Hemianopsia
o Blindness in R eye – cut right optic nerve
o Bitemporal hemianopsia – cut at optic chiasm – lose half the visual field (lateral of each eye)
o L homonymous hemianopsia – cut at optic tract – lose half visual field (entire L/R visual field)
o Upper quadrant hemianopsia – cut upper radiation – lose upper L/R ¼ quadrant of each eye
o Homonymous hemianopsia with macular sparing – cut striate cortex – sparing of the fovea with loss of ½ of L/R visual field
Visual Field Terminology
o Our vision is split into quadrants: upper right/left; lower right/left
o Binocular portion of visual field – visual field seen by two eyes
o Monocular portion of visual field – visual field seen by one eye
o Macula (center) of visual field is mapped onto a huge part (~1/2) of cortex known as occipital pole located at the very back end/tip of the brain
Visual Field Projections
o Upper visual field is mapped to the cortex below the calcarine sulcus
o Lower visual field is mapped to the cortex above the calcarine sulcus
o Left side of brain represent right visual field; Right side of brain represents left visual field
Neurons in V1 and Canonical Circuit
– responsible for recording information in a way that makes it more meaningful
25x as many neurons in V1 than LGN
o Projection neurons – pyramidal cells
o Local neurons – spiny stellate cells (excitatory); smooth stellate cells (inhibitory)
o Canonical Circuit
Magnocellular neurons synapse layer 4Calpha 4B 3/2 other cortical areas
• Also sends back to layer V or VI (goes back to LGN)
Parvocellular neurons synapse 4Cbeta 3/2 other cortical areas
• Also sends back to layer V or VI (goes back to LGN)
Cells in VI and Keys for Object Recognition
o Simple cells – small receptive field and first step in the brain where edges are distinguished; sends signal to complex cells
• Used to increase complexity (template matching: adds _ & | to make _|) (“AND”)
Sparse code – strong activation of a relatively small amount of neurons rather than activation of retina and ton of photoreceptors
Edges are great and efficient way to describe an object as long as we have simple cells
o Complex cells – larger receptive fields and translation tolerance (respond to edges in different positions)
Used to increase invariance (a max pooling operation: _ or | to make _) (“OR”)
o Two keys for objects recognition that occur in V1
An increase in future complexity (building more meaningful representations of the world)
Move toward invariant representation
Depth Perception
o Blending together of inputs in binocular cells where input is then from both eyes
o Critical for depth perception
• Clinical Correlation – Strabismus – eyes cannot be aligned correctly resulting in depth perception issues
o Treatment – correct it early on because there is a critical period for the visual system during which it learns how to do depth perception mapping
Binocular Cells and Disparity
o If you’re fixating on a certain point on one plane and something is placed in front of it, the brain compares where the new point is mapped on the two retinas to where the fixated point was mapped on the retinas
If the object is closer, it is mapped on a different location
Different cells are activated when fixating on an object and when there is disparity
o Zero disparity cell – no mismatch; activated when we fixate on a point
o Binocular disparity neuron – activated when we see an object on a closer/further plane (compared to the plane on which we have our fixated point)
Far or Near Cells
Principles of Cortical Organization (completeness and continuity)
o Completeness – the input space needs to be represented as completely as possible (“the sheet has to fill the box”)
o Continuity – neighboring neurons should have similar receptive fields
Minimal wiring constraint – most computations involve neurons with similar receptive fields; we want those close-by to avoid long connections
Mapping V2
– interweaved map of form, color, and depth
o Thick stripes are involved in processes with depth
o Thin stripes are involved in processes with color
o Pale stripes are involved in process with form
V2 Functions
o As you move up the visual field neurons become more sensitive to color and form; also become more invariant
o Continuous map of perceptual colors; tend to generalize the differentce in luminance
o Contour perception but NOT light; V1 responds to light but NOT contours
o Figure-ground segregation
V1 vs. V2 Neurons
V1 – neurons detect changes in luminance and color
Contain double opponent cells – cells that are less sesnsitive to illumination but more sensitive to color apple will be red whether you see it in morning or night/dark
V2 – detect contours and segregate between an object and its background
Does NOT respond to changes in luminance; instead are more invariant/constant
Streams of Visual Processing
– takes place after V2 when visual processing hierarchy diverges
Receptive fields increase and properties are increasingly complex
Retinotopy is progressively lost and topography becomes based on functional criteria
o Parietal/dorsal stream – tells us where something is in space via spatial and motion processing
Lesion – simultanagnosia (unable to perceive more than one object at a time), hemineglect (tendency to ignore the left side of objects and space)
o Temporal/ventral stream – tells us what an object is via object processing; significant input from parvocellular neurons
Lesion – visual form agnosias (unable to identify an object); associative agnosia (can’t name the object)
Summary: V1 and V2
o V1 – line orientation, spatial frequency, and component motion; basic color constancy, luminance, eye input, weak attentional effect
o V2 – perceptual contours/junctions/corners, perceptual colors/luminance/surfaces, binocular disparity, figure-ground segregation, component motion
Summary V4 and IT
o V4 – junctions, intermediate complex shapes, color, strong and specific attentional modulation, moderate invariance to size/position/shape, surface filling-in
o Infertemporal cortex (IT) – recognition of complex shapes, faces, color; binocular rivalry
Summary: MT, MST, LIP
o Medial Temporal (MT) (V5) – pattern motion
Neuron fires to “recognition” of a particular person whether its drawing/image/name
Lesion – cannot identify which direction dots are moving
o MST – smooth pursuit, optic flow (viewing rain while driving; following a fly move around)
o Lateral Intra-Parietal (LIP) – eye movement planning (saccades), strong attention/working memory effect, very weak object selectivity
Featured Units in Object Recognition
o Inferotemporal cortex (IT) – important in shape-tuning, unsupervised learning
Neurons recognize shape of cat vs. dog
o Pre-frontal cortex (PFC) – important in recognition-task specific tuning; supervised learning;
Required to determine that its actually a cat or dog
Send strong connections to IT neurons that “like” cats and weak connection to neurons that “like” dogs
Ventral Stream: HMAX Model of Object Recognition in Cortex
o HMAX Model of Object Recognition in Cortex – reflects understanding of how processing goes from simple cells complex cells simple cells complex cells repeatedly until projections move from V1 to the inferotemporal cortex (IT) and the prefrontal cortex (PFC)
Power of the brain comes from this cortical hierarchy
2 different operations to gradually increase invariance and selectivity
• Detect edges as input goes from LGN to simple cells
• Increase invariance (scale & lighting changes) as input goes simple complex
o Once you have representations in the IT, you can learn tasks by connecting the PFC to the IT
