Early visual processing for CG Early visual processing for CG Rich - - PowerPoint PPT Presentation

Early visual processing for CG Early visual processing for CG

Rich Clarke

Imagine we are building VR system. Money is no object:

• We are asked to specify the ultimate, fully immersive visual display device.

We are asked to specify the ultimate, fully immersive visual display device.

• What is the minimum that they must do, so that the resulting Virtual World is

utterly indistinguishable from reality?

• I.e. what are the hard limits on our perception of reality?

How to start to answer this question?

• Basic visual hardware: Visual anatomy & physiology

Basic visual hardware: Visual anatomy & physiology

• The fundamental unit of visual coding: Receptive fields
• Visual pathways: The split of information streams

p y p

• Brief aside: How do photoreceptors change EM radiation into neural

activity?

• How well can you really see? Resolution and acuity
• Seeing the sun and the stars: Brightness adaption
• Visual masking
• Colour perception

We’re going to keep in mind: The So What? Implications for Computer Graphics? Graphics?

The eye…

Basic visual hardware: Visual anatomy & physiology [Ferwerder] [ ]

POSTERIOR: ~Hexagonal mesh of ANTERIOR: Modelled as camera optics… POSTERIOR: ~Hexagonal mesh of photoreceptors…

Photoreceptors in the retina: Rods & Cones

Basic visual hardware: Visual anatomy & physiology

Pigmented layer Start of the ‘brain’…

LIGHT

To the optic nerve

LIGHT

Adapted from: [Ferwerder]

Basic visual hardware: Visual anatomy & physiology

Photoreceptors in the retina: Rods & Cones (2)

• Extremely sensitive to light
• Provide achromatic vision

Provide achromatic vision

• Work at low level (scotopic)

illumination

• Large receptive fields

g p

• Peak absorbance (sensitivity)

at ~500nm

• Less sensitive to light

Less sensitive to light

• Provide colour vision
• Work at high level (photopic)

illumination

• Three types:

‘B’ peak at 437nm, ‘G’ peak at 533nm, ‘R’ peak: 564nm Much smaller receptive fields

[Ferwerder]

Basic visual hardware: Visual anatomy & physiology

SO WHAT?

U d t di f “h d ” i i i ht i t ki d f i f ti th t

• Understanding of “hardware” gives insight into kinds of information that

can be coded

• Real VR system: focus resource on right areas

y g

The fundamental unit of visual coding: Receptive fields

A Cell’s Receptive Field

‘On centre’ cells

A Cell s Receptive Field

• Defined by spatially localised group of

photoreceptors serving some ganglion cell

‘Off centre’ cells Higher up in the brain (in V1): integrate simple cells:

photoreceptors serving some ganglion cell

• Location and quality of stimulus to which

the ganglion cell is responsive

cells simple cells: complex cells

the ganglion cell is responsive

• Opponency:

On centre and off centre Eye On centre and off centre Spectral as well as spatial: Red/Green Yellow Blue Absolute physical values lost: Absolute physical values lost:

[Ferwerder]

The fundamental unit of visual coding: Receptive Fields

SO WHAT?

B i b ildi bl k f i l ti

• Basic building blocks of visual perception
• Information about absolutes (both brightness and ‘colour’) lost –

contrast & context sensitivity only (c.f. illusions) y y ( )

• Brain ‘looks for’ fundamental structures that are/have been

behaviourally relevant in ontogenetic and phylogenetic history: fast ‘h d ’ iti ‘hardware’ recognition

Visual pathways: The split of information streams

From the eye to the brain

LGN 6 l

[Ferwerder]

LGN: 6 layers: Magnocellular layers – primary input from peripheral retina – non spectrally opponent ganglions large receptive fields

[Ferwerder]

non spectrally opponent ganglions, large receptive fields Parvocellular layers – primary input from the foveal region – spectrally opponent cells small receptive fields spectrally opponent cells, small receptive fields Area V1 (Visual Cortex): more complex cells (I.e. with more complex receptive fields) Deals with What? And Where? …Separately?

Visual pathways: The split of information streams

SO WHAT?

Eyes may be serving 2 relatively separate visual systems:

• Fast response, achromatic system, motion sensitive, low res.

(Magnocellular layers) Eyes may be serving 2 relatively separate visual systems: (Magnocellular layers)

• Slow response trichromatic system, motion insensitive, high res.

(Parvocellular layers)

How do photoreceptors change EM radiation into neural activity?

Rods & Cones: Outersegment: billions of light sensitive pigment molecules Outersegment: billions of light sensitive pigment molecules Molecules embedded in disks, stacked like pancakes Rods: Pigment is Rhodopsin γ

How well can you really see? Resolution and acuity

Physical limits on resolving power

Three things determine resolution: 1. Optical filtering, 2. receptor sampling, (and 3. receptive field organisation)

• 1. Real optical system: aberrations, diffraction at

entry aperture Resolution limited to 30arcsec

• 2. Photoreceptors sample retinal image -> neural

image representation. Spacing well matched to optics (Sampling theory) Visual acuity is a function of contrast sensitivity ~30sec Vernier (hyper)acuity: Ability to localise position of

• bjects – not a function of contrast sensitivity

Can detect misalignments of ~5sec Unknown exactly how it’s done

[Ferwerder]

How well can you really see? Resolution and acuity

SO WHAT?

• Obviously sets hard limit on how much detail required of a

Obviously sets hard limit on how much detail required of a display system

• VR systems not close to this for real-time display
• Vernier acuity plays an important role in the visibility of aliasing

artefacts in digital images – simple analysis of the visual system would predict that some artefacts should not be seen system would predict that some artefacts should not be seen (below the limit of supposed visual acuity)

Seeing the sun and the stars: Brightness adaption

Brightness adaption

How many orders of magnitude difference between the dimmest and the

g p

How many orders of magnitude difference between the dimmest and the brightest things we can see?

[Ferwerder]

Seeing the sun and the stars: Brightness adaption

Brightness adaption (2)

Three mechanisms

g p ( )

Mechanical

• Pupil dilation

Photochemical

• Bleaching & regeneration

Neural

• Changes in processing

Neural

• Changes in processing

Ch i C t t S iti it P tt A it C l P ti Changes in: Contrast Sensitivity: Pattern Acuity: Colour Perception

All: [Ferwerder]

Seeing the sun and the stars: Brightness adaption

The time course of adaption p

Purkeinje break

[Ferwerder]

Seeing the sun and the stars: Brightness adaption

SO WHAT?

• Most of the information or 'power' (in Fourier domain) of an

Most of the information or power (in Fourier domain) of an image is in brightness contrast

• Using 3 adaption mechanisms, able to see effectively over a

range of ~10 log units

• At different luminances contrast sensitivity, acuity, colour

perception changes markedly perception changes markedly

• Obvious implications for the design of a VR system (resource

allocation etc.)

Visual masking

Masking…

Defn: Visual Masking: Presence of 1 visual pattern changes visibility of another

g

Brightness: Top: 8 bit quantisation p q Bottom: 4 bit quantisation Experiments performed to determine parameters of visual masking Critical Band masking: Masking effect is greatest when frequency of the masking pattern is adjacent to the frequency of the i t image spectrum

[Ferwerder]

Visual masking

SO WHAT?

• Good example of how we do not see the world in veridicial

Good example of how we do not see the world in veridicial manner: masking changes way in which features perceived

• Contrast nonlinearity – implications for way that features of

world are coded by visual system

Colour perception [Purves & Lotto]

Colour perception [Purves & Lotto]

Colour perception

What is colour perception? How do we (efficiently) recreate it? p p ( y)

Relative stimulation of each cone type in your retina (RGB) in the context of some visual field visual field Different spectral distributions of light should be able to stimulate the photoreceptors identically: photoreceptors identically: Spectral distributions Receptor sensitivity SD a SD b SD b Distinct distributions that are perceived identically w.r.t some visual system - METAMERS

Colour Perception

SO WHAT?

• Sensible choice of some colour primaries should allow you to

Sensible choice of some colour primaries should allow you to re-create any visible colour simply (without recreating the whole C(λ) distribution

• Not quite as simple as that…but right primaries will produce a

colour gamut that covers most visible colours

Summary Summary…

• Understanding of “hardware”: insight into kinds of

information that can be coded R ti fi ld B i b ildi bl k f i l ti

• Receptive fields: Basic building blocks of visual perception
• Resolution of human visual system sets limit on how

much detail required much detail required

• At different luminances contrast sensitivity, acuity, colour

perception changes markedly perception changes markedly

• World is not seen “as it is”: Masking, colour context: need

to understand how scenes percieved to understand how scenes percieved

• Colour: Metamers, colour primaries & limitations of colour

gamuts g

Refs picture credits Refs, picture credits

Ref 1: [Ferwerder] Ferwerda J A (2001) Elements of Early Vision for Computer Ref 1: [Ferwerder] Ferwerda, J. A. (2001) Elements of Early Vision for Computer Graphics, IEEE Computer Graphics and Applications, 21(5), pp. 22-33. Ref 2: [Atkinson] R.C. Atkinson, ed., Steven’s Handbook of Experimental Psychology, 2nd ed John Wiley & Sons New York 1988 2nd ed., John Wiley & Sons, New York, 1988. Ref 3: [Purves & Lotto] www.lottolab.org, also D. Purves & R. Beau Lotto, Why we see what we do: An Empirical Theory of Vision, Sinauer Associates, 2003 Ref 7: [Sekuler & Blake] 7. R. Sekuler and R. Blake, Perception, McGraw-Hill, New York, 1994. Ref 15: [Spillman & Werner] L. Spillman and J.S. Werner, eds., Visual Perception: The Neurophysiological Foundations, Academic Press, San Diego, 1990. Ref 28: [Bollin & Mayer] M R Bolin and G M Meyer “A Frequency Based Ray Tracer ” Ref 28: [Bollin & Mayer] M.R. Bolin and G.M. Meyer, A Frequency Based Ray Tracer,

• Proc. Siggraph 95, ACM Press, New York, 1995, pp. 409-418.