constancy Flashcards
what is meant by perceptual constancy
the extent to which objects appear unchanging despite changes in the conditions of observing
what is size constancy
the tendency for objects to appear constant in size despite changes in the size of the retinal image caused by changes in viewing distance
what is shape constancy
the tendency for objects to appear constant in shape despite changes in the shape of the retinal image caused by changes in viewing angle
what is emmert’s law
the apparent size of the after image increases in direct proportion to viewing distance
the size of the retinal image of real objects depends on viewing distance
in contrast after images are fixed in size on the retina so apparent size changes with the distance it is projected (viewing distance)
how can we measure perceptual constancy
two modes: questions about sensations, or judgements of the real world
will produce different results
whichever method is used, less constancy will be obtained as contextual cues are reduced
participants make systematic errors in judgements about an image that matched their retinal image
tend towards the properties of the object rather than the properties of the retinal image
what did Thouless (1931) mean by phenomenal regression to the real world
when making judgements about an image people make systematic errors
what is colour constancy
the tendency for objects to appear unchanging in colour despite changes in the spectral composition of the illuminant
how is light reflected and absorbed from the surface of an object
as spectral energy as a function of wavelength
thus reflected light = e(𝞴).
what is reflected light
illumination l(𝞴) x surface reflectance R(𝞴)..
how can we mathematically represent cone signals
SML cone spectral sensitivity x surface reflectance x illumination (integrated over all wavelengths)
S = ∫ s(λ)R(λ)I(λ)dλ
M = ∫ m(λ)R(λ)I(λ)dλ
L = ∫ l(λ)R(λ)I(λ)dλ
note we perceive object properties rather than the properties of the retinal image
what is the effect of a change in illumination
changes in illumination change the reflected light and change the cone signals
despite this we tend to perceive a colour that is associated with surface reflectance thus maintaining its constant appearance
what are the three components of colour constancy
the calculations required to undo the effect of an illuminant change
the parameters of the calculations, determined by the image
the perceptual apparatus where these transformations are implemented
what happens to cone signals when the light changes
scaling factors are required to compensate for illuminant changes
different illuminant changes produce different scaling factors
the perceptual system can only draw on prior knowledge of scenes it has already viewed
what is von kries’ coefficient rule (1878-1905)
light adaption influences colour appearance by
turning the signal from each cone class up or down by a multiplicative coefficient
these changes within each cone class are independent at each retinal location such that signals from M S cones do not influence L cones
for each cone class the gains are set in inverse proportion to the spatial average of the photon caught by cones of the same class
the size and shape of a photon varies with the photon energy or wavelength
perceived colour is determined by the ratios of the scaled cone signals (rather than raw cone signals)
can the effect of spectral changes in the illuminant be summarised by a multiplicative scaling of the cone signals
This is not mathematically true for all lights, surfaces and photoreceptor sensitivities.
This is true for the sets of real lights and surfaces likely to be encountered by a visual system with particular photoreceptors.
There is regularity in the physical world that the perceptual system can exploit.
what is the ives transform (1912)
First mechanism for constancy under an illuminant change
Demonstrated that the multiplicative factors which transform the coordinates of incandescent carbon illumination to those of a reference illuminant, also transform the coordinates of surfaces to approximately their coordinates under the reference illuminant.
BUT in practice the visual system rarely has direct access to the cone-coordinates of the illuminant and instead the illuminant has to be estimated from cues in a complex scene.
This could operate by assuming that the brightest illuminant is white.
what is the argument that we assume ‘the brightest is white’
the brightest surface in a scene is likely to reflect all wavelengths in the illuminant spectrum without alteration, appearing white
we are good at identifying white surfaces
this only uses one data - much more robust to have many
what it the global mean argument
global mean = mean of all data points
if constant, can be used to estimate the gradient of the line
however, normalising the global mean will only provide constancy if the surfaces in a scene remain constant in their average properties
what is the grey world hypothesis?
assumes that the average reflectance from a group of surfaces is neutral and consistently reflects the illuminant
normalising the mean for a green biassed collection of surfaces under sunlight overcorrects the L cone signal
normalising the mean of a red biassed collection of surfaces under skylight under corrects the L cone signal
we wrongly attribute the bias of these surfaces to the illuminant resulting in under or over corrections
GWH says that the set of surfaces stay constant and on average are grey.
BUT statistically this does not work as we do not average things out to be grey.
Better to argue for a constant world hypothesis or a set of biases that over a particular amount of time might be constant.
what is the local mean
Kries suggested that the signal in each cone class was normalised to the local mean for that cone class
this achieves constancy but eliminates differences between surfaces
normalising to the local mean allows corrected sunlight images to be the same as corrected skylight images, but you lose the structure and variation across the original scene
normalising to the global mean is a more useful operation which may provide an estimate of the illuminant
what is Smithson & Zaidi’s 2004 temporal mean
in an experiment of perceived colour constancy
participants were presented with a series of frames with a central square every 1.5s, and were asked to judge the colour of the square
illumination was manipulated such that the entire scene was illuminated by skylight or sunlight, or the frame and square were illuminated differently
they found that when the entire scene was illuminated uniformly, constancy was high
but when the test patch was illuminated locally by a different cast light to the background, its colour was perceived correctly to the local illumination (rather than the background illumination)
We normally make judgements about colour constancy relative to the surroundings and from a single view of a test patch our visual system cannot deduce that it is rendered in a different light from the surround
this suggests that the local illuminant was estimated over successive trials
eye movements operate in natural scenes to help average out different biasses in different parts of a scene and extract a reliable temporal estimate of the average properties of the illuminant
thus the global mean does not have a big effect on the colour appearance of the test patch
rather than a spatial average we consider the recent history of the surfaces we’ve seen
what is Kraft & Brainard’s 1999 multiple cue proposal
in an experiment of perceived colour constancy one perceptual cue was systematically silenced to infer the relative influence each one has on colour constancy
tested the contribution of three classic hypotheses:
local adaption
adaption to the spatial mean
adaption to the most intense image region
actual illuminated surfaces were used and independently manipulated
observers used computer controls to make an achromatic scene which informs the experimenters of what the observer perceives the illuminant light to be (white surfaces perfectly reflect the illuminant light)
local surrounds were equated to the test patch so that they gave no information about the different illuminants (grey cardboard for neutral illumination, blue cardboard for orange-red light)
equated maximum flux so that the brightest area is the same colour in both the neutral illuminant and yellow light to silence the brightest is white cue (frame of yellow cardboard, neutral illumination; frame of magenta cardboard, yellow light)
found that constancy gets worse as the number of available cues decreases (i.e. judgements are closer to the retinal stimulus)
there is individual variation in the way that multiple cues are used suggesting neither of the classic theories are sufficient
multiple cues may be used in combination and other uncontrolled cues may be playing a role as performance never falls to failure of constancy
how can we explain ‘the dress’
individual differences in illuminant estimation
the photo does not have many cues so constancy breaks down somewhat as it is difficult to correct for the illuminant
we focus on irrelevant cues giving false information
we can manipulate perception by manipulating the context its seen in
need to deduce the relative contributions of the illuminant colour spectrum and the surface reflection to the overall perception
Either the dress is in bluish light/ shadow.
In this scenario the only reflectance that can be multiplied is flare and therefore is a white stripe.
Or the dress is in yellow light
In this scenario the light must be hitting a blue surface that reflects a lot of short wavelength but not long wavelength light and is therefore a blue stripe.
Depending on which judgements are made the perception of the dress will change.
how can we measure colour constancy
matching a standard display
matching to an internal standard
colour naming
matching to a standard display: simultaneous asymmetric matching (Adjust the colour of the central square in the righthand display to match the colour of the central
square in the left-hand display.), successive asymmetric matching (Adjust the colour of the central square in the second
display to match the colour of the central square in
the first display), haploscopic matching (Adjust the colour of the central square in the righthand display to match the colour of the central
square in the left-hand display)
matching to an internal standard: achromatic setting (Adjust the colour of the central square so that it looks
grey or colourless.), colour boundaries (Does the central square appear reddish or greenish?
Does the central square appear yellowish or bluish?)
colour naming (What colour is the central square?)
why does it matter what question you ask
Arend & Reeves (1986) presented a clear demonstration of the influence of instructions in a colourmatching task.
When observers were asked to make a match to ‘look as if it were cut from the same piece of
paper’, they showed relatively good constancy compared with conditions where they were asked
to match ‘hue and saturation’.
The appearance-based constancy obtained in the second case is a demonstration of
“phenomenal regression to the real object” (Thouless 1931).
what is the neural implementation of colour constancy
scaling of cone signals supports a high level of constancy and these transformations should happen from the level of the photoreceptor to later signals
summarise colour constancy
Perceptual constancy is all about recovering the properties of objects in the world, and not the properties of the retinal image.
In measuring perceptual constancy, the result can be influenced by the way the question is asked. However, it is clear that with fewer cues to the viewing circumstances, constancy is reduced.
In understanding constancy, we must consider three components of the problem:
What calculations are required to “undo” the effect of the change in viewing circumstances?
How are the parameters of the calculations determined by the image?
Where in our perceptual apparatus are these transformations implemented?
For colour constancy, a multiplicative change in cone signals provides a good approximate solution to the problem. The multiplicative factors may be set by the mean (over an appropriate set of elements) or by the brightest elements.
It is likely that several image features are used in combination to set the correction, and implemented at multiple stages in the visual system.
