Econstudentlog

Perception (I)

Here’s my short goodreads review of the book. In this post I’ll include some observations and links related to the first half of the book’s coverage.

“Since the 1960s, there have been many attempts to model the perceptual processes using computer algorithms, and the most influential figure of the last forty years has been David Marr, working at MIT. […] Marr and his colleagues were responsible for developing detailed algorithms for extracting (i) low-level information about the location of contours in the visual image, (ii) the motion of those contours, and (iii) the 3-D structure of objects in the world from binocular disparities and optic flow. In addition, one of his lasting achievements was to encourage researchers to be more rigorous in the way that perceptual tasks are described, analysed, and formulated and to use computer models to test the predictions of those models against human performance. […] Over the past fifteen years, many researchers in the field of perception have characterized perception as a Bayesian process […] According to Bayesian theory, what we perceive is a consequence of probabilistic processes that depend on the likelihood of certain events occurring in the particular world we live in. Moreover, most Bayesian models of perceptual processes assume that there is noise in the sensory signals and the amount of noise affects the reliability of those signals – the more noise, the less reliable the signal. Over the past fifteen years, Bayes theory has been used extensively to model the interaction between different discrepant cues, such as binocular disparity and texture gradients to specify the slant of an inclined surface.”

“All surfaces have the property of reflectance — that is, the extent to which they reflect (rather than absorb) the incident illumination — and those reflectances can vary between 0 per cent and 100 per cent. Surfaces can also be selective in the particular wavelengths they reflect or absorb. Our colour vision depends on these selective reflectance properties […]. Reflectance characteristics describe the physical properties of surfaces. The lightness of a surface refers to a perceptual judgement of a surface’s reflectance characteristic — whether it appears as black or white or some grey level in between. Note that we are talking about the perception of lightness — rather than brightness — which refers to our estimate of how much light is coming from a particular surface or is emitted by a source of illumination. The perception of surface lightness is one of the most fundamental perceptual abilities because it allows us not only to differentiate one surface from another but also to identify the real-world properties of a particular surface. Many textbooks start with the observation that lightness perception is a difficult task because the amount of light reflected from a particular surface depends on both the reflectance characteristic of the surface and the intensity of the incident illumination. For example, a piece of black paper under high illumination will reflect back more light to the eye than a piece of white paper under dim illumination. As a consequence, lightness constancy — the ability to correctly judge the lightness of a surface under different illumination conditions — is often considered to be an ‘achievement’ of the perceptual system. […] The alternative starting point for understanding lightness perception is to ask whether there is something that remains constant or invariant in the patterns of light reaching the eye with changes of illumination. In this case, it is the relative amount of light reflected off different surfaces. Consider two surfaces that have different reflectances—two shades of grey. The actual amount of light reflected off each of the surfaces will vary with changes in the illumination but the relative amount of light reflected off the two surfaces remains the same. This shows that lightness perception is necessarily a spatial task and hence a task that cannot be solved by considering one particular surface alone. Note that the relative amount of light reflected off different surfaces does not tell us about the absolute lightnesses of different surfaces—only their relative lightnesses […] Can our perception of lightness be fooled? Yes, of course it can and the ways in which we make mistakes in our perception of the lightnesses of surfaces can tell us much about the characteristics of the underlying processes.”

“From a survival point of view, the ability to differentiate objects and surfaces in the world by their ‘colours’ (spectral reflectance characteristics) can be extremely useful […] Most species of mammals, birds, fish, and insects possess several different types of receptor, each of which has a a different spectral sensitivity function […] having two types of receptor with different spectral sensitivities is the minimum necessary for colour vision. This is referred to as dicromacy and the majority of mammals are dichromats with the exception of the old world monkeys and humans. […] The only difference between lightness and colour perception is that in the latter case we have to consider the way a surface selectively reflects (and absorbs) different wavelengths, rather than just a surface’s average reflectance over all wavelengths. […] The similarities between the tasks of extracting lightness and colour information mean that we can ask a similar question about colour perception [as we did about lightness perception] – what is the invariant information that could specify the reflectance characteristic of a surface? […] The information that is invariant under changes of spectral illumination is the relative amounts of long, medium, and short wavelength light reaching our eyes from different surfaces in the scene. […] the successful identification and discrimination of coloured surfaces is dependent on making spatial comparisons between the amounts of short, medium, and long wavelength light reaching our eyes from different surfaces. As with lightness perception, colour perception is necessarily a spatial task. It follows that if a scene is illuminated by the light of just a single wavelength, the appropriate spatial comparisons cannot be made. This can be demonstrated by illuminating a real-world scene containing many different coloured objects with yellow, sodium light that contains only a single wavelength. All objects, whatever their ‘colours’, will only reflect back to the eye different intensities of that sodium light and hence there will only be absolute but no relative differences between the short, medium, and long wavelength lightness records. There is a similar, but less dramatic, effect on our perception of colour when the spectral characteristics of the illumination are restricted to just a few wavelengths, as is the case with fluorescent lighting.”

“Consider a single receptor mechanism, such as a rod receptor in the human visual system, that responds to a limited range of wavelengths—referred to as the receptor’s spectral sensitivity function […]. This hypothetical receptor is more sensitive to some wavelengths (around 550 nm) than others and we might be tempted to think that a single type of receptor could provide information about the wavelength of the light reaching the receptor. This is not the case, however, because an increase or decrease in the response of that receptor could be due to either a change in the wavelength or an increase or decrease in the amount of light reaching the receptor. In other words, the output of a given receptor or receptor type perfectly confounds changes in wavelength with changes in intensity because it has only one way of responding — that is, more or less. This is Rushton’s Principle of Univariance — there is only one way of varying or one degree of freedom. […] On the other hand, if we consider a visual system with two different receptor types, one more sensitive to longer wavelengths (L) and the other more sensitive to shorter wavelengths (S), there are two degrees of freedom in the system and thus the possibility of signalling our two independent variables — wavelength and intensity […] it is quite possible to have a colour visual system that is based on just two receptor types. Such a colour visual system is referred to as dichromatic.”

“So why is the human visual system trichromatic? The answer can be found in a phenomenon known as metamerism. So far, we have restricted our discussion to the effect of a single wavelength on our dichromatic visual system: for example, a single wavelength of around 550 nm that stimulated both the long and short receptor types about equally […]. But what would happen if we stimulated our dichromatic system with light of two different wavelengths at the same time — one long wavelength and one short wavelength? With a suitable choice of wavelengths, this combination of wavelengths would also have the effect of stimulating the two receptor types about equally […] As a consequence, the output of the system […] with this particular mixture of wavelengths would be indistinguishable from that created by the single wavelength of 550 nm. These two indistinguishable stimulus situations are referred to as metamers and a little thought shows that there would be many thousands of combinations of wavelength that produce the same activity […] in a dichromatic visual system. As a consequence, all these different combinations of wavelengths would be indistinguishable to a dichromatic observer, even though they were produced by very different combinations of wavelengths. […] Is there any way of avoiding the problem of metamerism? The answer is no but we can make things better. If a visual system had three receptor types rather than two, then many of the combinations of wavelengths that produce an identical pattern of activity in two of the mechanisms (L and S) would create a different amount of activity in our third receptor type (M) that is maximally sensitive to medium wavelengths. Hence the number of indistinguishable metameric matches would be significantly reduced but they would never be eliminated. Using the same logic, it follows that a further increase in the number of receptor types (beyond three) would reduce the problem of metamerism even more […]. There would, however, also be a cost. Having more distinct receptor types in a finite-sized retina would increase the average spacing between the receptors of the same type and thus make our acuity for fine detail significantly poorer. There are many species, such as dragonflies, with more than three receptor types in their eyes but the larger number of receptor types typically serves to increase the range of wavelengths to which the animal is sensitive into the infra-red or ultra-violet parts of the spectrum, rather than to reduce the number of metamers. […] the sensitivity of the short wavelength receptors in the human eye only extends to ~540 nm — the S receptors are insensitive to longer wavelengths. This means that human colour vision is effectively dichromatic for combinations of wavelengths above 540 nm. In addition, there are no short wavelength cones in the central fovea of the human retina, which means that we are also dichromatic in the central part of our visual field. The fact that we are unaware of this lack of colour vision is probably due to the fact that our eyes are constantly moving. […] It is […] important to appreciate that the description of the human colour visual system as trichromatic is not a description of the number of different receptor types in the retina – it is a property of the whole visual system.”

“Recent research has shown that although the majority of humans are trichromatic there can be significant differences in the precise matches that individuals make when matching colour patches […] the absence of one receptor type will result in a greater number of colour confusions than normal and this does have a significant effect on an observer’s colour vision. Protanopia is the absence of long wavelength receptors, deuteranopia the absence of medium wavelength receptors, and tritanopia the absence of short wavelength receptors. These three conditions are often described as ‘colour blindness’ but this is a misnomer. We are all colour blind to some extent because we all suffer from colour metamerism and fail to make discriminations that would be very apparent to any biological or machine vision system with a greater number of receptor types. For example, most stomatopod crustaceans (mantis shrimps) have twelve different visual pigments and they also have the ability to detect both linear and circularly polarized light. What I find interesting is that we believe, as trichromats, that we have the ability to discriminate all the possible shades of colour (reflectance characteristics) that exist in our world. […] we are typically unaware of the limitations of our visual systems because we have no way of comparing what we see normally with what would be seen by a ‘better’ visual system.”

“We take it for granted that we are able to segregate the visual input into separate objects and distinguish objects from their backgrounds and we rarely make mistakes except under impoverished conditions. How is this possible? In many cases, the boundaries of objects are defined by changes of luminance and colour and these changes allow us to separate or segregate an object from its background. But luminance and colour changes are also present in the textured surfaces of many objects and therefore we need to ask how it is that our visual system does not mistake these luminance and colour changes for the boundaries of objects. One answer is that object boundaries have special characteristics. In our world, most objects and surfaces are opaque and hence they occlude (cover) the surface of the background. As a consequence, the contours of the background surface typically end—they are ‘terminated’—at the boundary of the occluding object or surface. Quite often, the occluded contours of the background are also revealed at the opposite side of the occluding surface because they are physically continuous. […] The impression of occlusion is enhanced if the occluded contours contain a range of different lengths, widths, and orientations. In the natural world, many animals use colour and texture to camouflage their boundaries as well as to fool potential predators about their identity. […] There is an additional source of information — relative motion — that can be used to segregate a visual scene into objects and their backgrounds and to break any camouflage that might exist in a static view. A moving, opaque object will progressively occlude and dis-occlude (reveal) the background surface so that even a well-camouflaged, moving animal will give away its location. Hence it is not surprising that a very common and successful strategy of many animals is to freeze in order not to be seen. Unless the predator has a sophisticated visual system to break the pattern or colour camouflage, the prey will remain invisible.”

Some links:

Perception.
Ames room. Inverse problem in optics.
Hermann von Helmholtz. Richard Gregory. Irvin Rock. James Gibson. David Marr. Ewald Hering.
Optical flow.
La dioptrique.
Necker cube. Rubin’s vase.
Perceptual constancy. Texture gradient.
Ambient optic array.
Affordance.
Luminance.
Checker shadow illusion.
Shape from shading/Photometric stereo.
Colour vision. Colour constancy. Retinex model.
Cognitive neuroscience of visual object recognition.
Motion perception.
Horace Barlow. Bernhard Hassenstein. Werner E. Reichardt. Sigmund Exner. Jan Evangelista Purkyně.
Phi phenomenon.
Motion aftereffect.
Induced motion.

October 14, 2018 - Posted by | Biology, Books, Ophthalmology, Physics, Psychology

No comments yet.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: