Visual Perception & Cognition PDF

Levels of visual representation and processing We are able to represent images even if it is partially occluded Theory of vision: eye is like a camera (mental image is created) Mental images are analogical Mental image requires an internal eye that requires another internal eye (etc.) Image m...

Levels of visual representation and processing We are able to represent images even if it is partially occluded Theory of vision: eye is like a camera (mental image is created) Mental images are analogical Mental image requires an internal eye that requires another internal eye (etc.) Image makes no distinction between objects, figure, ground An analysis is still required **Images need to be coded we compute representations Representations are stable, requiring/allows for updating Allow for reference thinking, and action The image from the world is not perfect (holes in the image) Optic disk/blindspot: no photoreceptors because that is where the optic nerve is The brain (visual system compensates) Image we get requires a lot of transformations (computations) Low-level processes From Eye to V1 & Beyond Reception: absorption of physical energy by the receptors Transduction: conversion of physical energy into an electrochemical pattern for the neurons Coding: 1-to-1 correspondence between aspects of the physical stimulus and aspects of the resultant nervous system activity Representations & Processes: what we do with what we encode from the world How vision affords knowledge and action Physical Implementation (Implementation level) Easy to explain by Connectionnist Architecture Fovea: concentration of cones in the retina High [ ] of cones / no rods at fovea Higher [ ] of rods away from fovea Lower [ ] of cones away from fovea Rods: allow to see in low-light conditions Very sensitive No contribution to colour vision Cone: provide high-acuity colour vision in bright (daylight) conditions Less sensitive Retina: where the photoreceptor cells responsible for sensory transduction are located RECEPTIVE FIELDS Receptive field [of a neuron]: area on the retina for which stimuli affect that neuron's firing rate Each ganglion cell axon in the optic nerve will respond to a specific location on the retina ◦ P/midget: small receptive fields, high acuity Work best in high luminance situations Sustained firing Information about the contrast ◦ M/parasol: large receptive fields, low acuity Works best in low luminance situations Burst firing Information about change over time of an image LATERAL INHIBITION Lateral inhibition: Light falling on photoreceptors surrounding the one in the middle will inhibit its response leads to annular/ring-shaped receptive fields Arrangement for an "on-center" ganglion cell "ON-center" ganglion cell ◦ Excited by light falling on the center of the receptive field ◦ Inhibited by light falling in surrounding areas ◦ Trigger the cones in surrounding space that will send inhibitory signal to horizontal cells ◦ Sensitive to size of illumination = higher response at a specific size and smaller response if smaller/bigger "OFF-center" ganglion cell ◦ Inhibited by light falling on the center of the receptive field ◦ Trigger the cones in the centre that will send inhibitory signal to horizontal cells ◦ Excited by light falling on the surrounding area ROUTE OF VISUAL SIGNALS From retina to the primary visual cortex (V1) through lateral geniculate nucleus (LGN) Optic nerves from each eye meet at the optic chiasm Left visual field of each eye go to the right LGN Right " " " " " " to the left LGN LGN Part of the thalamus Arranged in layers 1 & 2: magnocellular layer from M (parasol) ganglion cells 3 - 6: parvocellular layers from P (midget) ganglion cells 1,4 & 6: receive input from the contralateral eye 2, 3 & 5: receive input from the ipsilateral eye Koniocellular layers may be involved in colour vision Get input from K type ganglion cells VISUAL HEMIFIELDS Increasing overlap at higher fibre projections: V1 = 1 degree of visual field overlap V4 = 5 degree of visual field overlap IT = almost full (60 degrees) visual field overlap Low-level processing of visual information Primary Visual Cortex (V1) TOPOGRAPHIC MAPPING Retinotopic mapping: relative position of objects within the 2 halves of our visual field Maintained in V1 Cortical magnification: relative amount of V1 dedicated to processing visual information from the fovea relative size of the retinal image More neurons of V1 needed to process central vision than peripheral vision FEATURE DETECTORS IN V1 1. Simple cells: respond best to bars of light (or dark but not both) with a particular orientation and location on the retina Strong response to bars of a specific orientation Less response to bars of similar orientation Contrast has no effect the orientation Contrast affects the magnitude of response 2. Complex cells: respond equally well to bars of light and dark with a particular orientation anywhere within their receptive field Orientation-tuned response but location of stimulus is not important Highly responsive to moving bars within their receptive field Receptive field is usually larger than simple cells 3. End-stopped cells: are selective for the length of the bar of light they are stimulated with In V1 V4 = colour, edges, curvatures, contours Multiple simple cells Slightly different orientation selectivity Synapse onto a neuron in V4 Form a contour-selective receptive field LOC (lateral occipital cortex) & IT (inferotemportal cortex) = objects, faces, places MT = motion SELECTIVITY OF CELLS IN V1 Achromatopsia: impaired colour perception (V4) Akinetopsia: motion impaired (V5, MT) Specialization is not absolute: neuronal pathways are interconnected Different groups of cells at different areas respond to different types of stimuli evidence for modular coding/representations: brain areas responsible to specific stimuli Limitations of comparative studies: Similar areas in humans and monkeys may have different functions Similar functions may be modulated by different areas PATHWAYS Dorsal stream (Where or how pathway) V1 > V2 > MT > parietal lobe Represents properties that relate to objects' motion or location Guided action Ventral stream (What pathway) V1 > V2 > V4 > inferotemporal cortex Represents properties that represent objects' identities Shape, colour High-level processes Perception Perception: study of organization of the external world as encoded by our senses How do we encode (recognize, categorize, act upon) the world Like a reflex: computational mechanisms that act like they are “smart” but are “dumb”  Has its own rules WHY SMART? Pre-encoded rules Knows a lot about how to encode properties the world How to interpret the world i.e. an object that expands in your retina = the object is coming towards you Has rules and representations Gestalt vision rules Can extract shape and motion from shading Convexity The visual system relies on regularities of the world Light tend to shine from above (shade above = concave; shade below = convex) WHY DUMB? Makes assumptions Large part of it operates without relaying on higher cognitive mechanisms Does a job automatically, like a reflex It can make mistakes Persistence of visual illusions: knowing it is an illusion does not alter your perception Inverse projection problem: The same object can project different images onto the retina but different images can also project the same image onto the retina Levels of Analysis in Vision LOW LEVEL Product of light intensities across an image or visual field (2D) Low level computations produce basic surface layout information (edges, discontinuity, shape) INTERMEDIATE LEVEL Surface and 3D information Computations operate over surface discontinuity and produce viewer-centred 3D information 21/2 D representation, only 3D from your own point of view HIGH LEVEL Object, face, and scene representation object-entered computations Bottom Up/Top Down Visual chart: Detectors: different subsystems working in parallel Visual perception (low level) Low level computations Integration between different types of representations Visual Cognition (high level) Object, face, scene representations Semantic/Conceptional system (and others) Attach meaning, knowledge, and memory