Color perception

[From Bruce Abbott (2001.05.19.1730 EST)]

Bill Powers (2001.05.19.1142 MDT) --

Bruce Abbott (2001.05.18.1800 EST)

Bill Powers (2001.05.18.1445 MDT)

Sort of like Bruce Abbott and his "precisely tuned color
filters" that he invents and then rejects (in favor of his own view, oddly
enough).

Bill, if you are going to make comments like this, the honorable thing to do
is to support them with some sort of reasoned argument. I trust that one is
coming.

"Argument?" I was just echoing what you said a couple of days ago about
color vision, and what you attributed to me (finely-tuned color filters).

You should go back and re-read that post, if you still have it. I didn't
attribute that model to you at all! I proposed that first-order and
second-order perceptions both might be apprehened consciously (when they
are) as sensations. I suggested that some first-order signals might feed
unchanged into those high-level sensory analyzers whose activities give rise
to conscious perception and yield, not dimensionless intensities, but
particular sensations varying in intensity with the neural current. Others,
I suggested, might go no further than the inputs to second-order functions,
leaving only second-order signals going to the high-level analyzers. To
illustrate the latter, I presented a scheme for color perception where the
outputs of the three primary color receptors are combined in various
proportions to yield all the perceptible colors. I contrasted the
efficiency of this solution to that of an alternative in which every
perceivable color is represented by a separate, narrow-band frequency
sensor. Neither example was intended to be more than an illustration of
principle.

There are many kinds of perceptual input functions that could account for
color vision. One that I particularly like is Edwin Land's proposal, in
which color sensations are functions of long and short wavelength
intensities and an average of intensities over all receptors.

In using the color-vision example to illustrate a second-order perceptual
function, I was not suggesting that this is the actual mechanism of color
perception found in the human brain. I did see it as an example that would
be easy to understand and sensible in view of both the existence of three
detectors broadly tuned to three different freuencies and our experiences
resulting from mixing red, green, and blue lights. After all, only those
three colors are actually present in a color TV's output and yet we perceive
almost the entire rainbow of colors in the screen. It also meets the formal
requirements of a second-order PCT-style perceptual function.

However, I am aware that this simple scheme, elegant though it is,
represents a great simplification of the actual human color-perception
system. Even at the level of the retina, there are bipolar neurons whose
output signals are complex functions of the outputs of the three types of
color receptor. Some of these bipolar cells function very much like the
opponent-cells envisioned by Ewald Hering in the late 19th century, who
based his theory of "opponent process" theory of color perception partly on
observations of various forms of abnormalities in color vision. An example
is the "red +, green -" output: light in the low-frequency region (perceived
as red when present alone) excites the cell's output while light in the
middle-frequency region (perceived as green when present alone) inhibits the
cell's output. (Things get even more complex at the ganglion-cell level,
the next stage in color processing).

I understand that there has been quite a bit of progress recently in working
out the actual physiological mechanisms involved in color processing (and
visual processing in general); I'd really like to become much more familiar
with this literature. If anyone tuned into CSGnet is familiar with these
recent advances, it would be nice get a synopsis of them.

I have a fairly recent, relatively low-level book on visual perception
[Hoffman, Donald D. (1998). _Visual intelligence_. New York: Norton] that
hints at the sophistication of the visual system. As one example, the color
perceived in a swatch of color arranged alongside many others is a function
not only of the frequencies of light being reflected by the swatch, but also
on the frequencies being reflected by other swatches (color contrast) and on
what the visual system "takes" to be the _color-balance of the light shining
on the surface_ (!). (The latter process attempts to separate color as a
property ascribed to the object from color as a property of the illumination
and not of the object.) Another example is the perception of illusory
colored objects, where the brain has painted in a nonexistent object as a
way of explaining certain regularities seen in the objects actually present.
By covering the objects, you can demonstrate for yourself that the illusory
colored object is not really there, as it disappears when the real objects
are removed.

I don't know the details of Land's theory. Does it begin with the three
types of color receptor cells in the retina (as Helmholtz had inferred were
present)? I don't know whether the presence of these three tuned receptors
had been confirmed in Land's day; if not, then it is possible he worked out
a theory that did not start with these three frequencies as primary.

Bruce A.

[From Bruce Abbott (2000.05.19.2220 EST)]

Ah, a nice discussion of color vision, including Land's Retinex theory (and
an explanation of his demo) appears on the web at:

http://webvision.med.utah.edu/Color.html

This is probably as up-to-date a review of current thinking and data in this
area of research as one could hope for . . .

Bruce A.

[From Bill Powers (2001.05.20.1623 MDT)]

Bruce Abbott (2000.05.19.2220 EST)]

Ah, a nice discussion of color vision, including Land's Retinex theory (and
an explanation of his demo) appears on the web at:

http://webvision.med.utah.edu/Color.html

This is probably as up-to-date a review of current thinking and data in this
area of research as one could hope for . . .

Good. I downloaded the Gouras paper and read it. My first impression was
the same as yours, but on re-reading I realized that there is still a long
way to go. What we really would like would be an analysis that says "When
the input intensities over the retina are such and such at the following
wavelengths, a color of C will be seen at coordinates X,Y." That still lies
a bit in the future.

I did get a bit tired of the attribution of intelligence to "Nature" and
the explanations of why evolution made things work the way they do. That's
just filler material. It may be true that color discrimination allows us to
see things that intensity contrasts could not reveal, but so what? It also
might be that color-blindness would help us penetrate camoflage, as the
military discovered long ago. If we didn't see in color, we could explain
why that's best, too. After all, color contrasts can be unimportant surface
features not reflecting, and indeed disguising, forms. Maybe animals that
are color-blind are protected from seeing so much information that their
small brains would be overloaded. No matter what the case is, you can
"explain" it this way.

It's not at all clear where various stages of visual signal processing take
place. If you find a signal at a given location that seems to correspond to
some visual attribute, you can be sure that the neural functions that
derived that attribute are located _lower_ in the nervous system. I had
this problem with the Hubel and Weisel findings, too: if a cell fires when
a vertical line appears, is that cell the detector of vertical lines, or it
is being excited by the output of some lower cell-assembly that is
organized to create maximum signal for that direction? The literature about
such perceptual vectors has never been clear about where the relevant
computations are taking place.

Intensity signals are signals that correspond to the intensity of
stimulation: in the visual or auditory channels that would correspond to
input energy flux. Sensations then would be functions of sets of these
intensity signals. Different receptors -- long and short-wavelength cones,
for example -- generate signals which represent the sensors' response to
energy fluxes in different wavelength bands. But these are still intensity
signals until, as Goulas says, "the brain compares two signals to
distinguish color." I would not use the term "compare" here, however:
"combine" would be a more appropriate general term. A color signal is some
function (not necessarily just A minus B)of a set of intensity signals from
different receptors, the nature of the function remaining to be discovered.
I suppose it would be something like a weighted sum, although there are
evidently feedback effects from many surrounding cells that scale the
measurements relative to a global average color treated as neutral or gray.

There is a possibility that different colors over some range might be
represented as the magnitude of a single color signal. This would appear to
be the case when only one single color at a time can be detected in a given
place. This would imply that there is only one signal representing color.
Suppose that red were represented as a neural frequency of 10 Hz, orange as
100 Hz, and yellow as 500 Hz. This situation could occur if a single input
function were producing a single output signal as a weighted sum of
intensity signals coming from long and short wavelength receptors. To make
this work, there would have to be a local feedback loop, like an automatic
gain control, that maintains the total of all signals at this level at some
constant magnitude (as Land's theory and experiments strongly suggest).
This would mean that a given signal magnitude would have meaning only
relative to all other color signals at the same level, and that signal
magnitude would no longer represent the intensity of input stimulation. In
fact, this would make it possible for a signal magnitude to change with
_wavelength_ (or better, _color_) while being nearly invariant with respect
to input intensity.

A very similar arrangement could exist in the auditory channels. Signals
that carry intensity (loudness) information at the lowest level give rise
to signals that indicate pitch at the second level, where we can guess that
pitch-sensations become invariant with respect to loudness due to
gain-control effects. Now the magnitude of the neural signal could indicate
pitch at the second level, whereas essentially identical signals would
indicate loudness at the first level.

The auditory channels are very confusing because the neural signals
representing sounds have frequencies in a range that overlaps the range of
physical pressure oscillations being sensed. It's much easier to understand
this relationship in the visual modality, where the input oscillations (of
electromagnetic fields) are something like a trillion times higher in
frequency than the spike frequencies that represent them. In the visual
field, we can see that neural frequencies are simply the common measure of
all variables, whatever their physical significance. So intensities are
represented as neural frequencies, and at the next level, colors are
represented as neural frequencies, and I presume at the next level still,
configurations, forms are represented as neural frequencies, and so on.

The question of conscious perception of these visual attributes is confused
in part by the uncertainty of just where the computations take place
(mentioned above), and also by the assumption that for conscious perception
to occur, signals must reach the cerebral cortex. This assumption would be
hard to verify -- it's simply supposed that since "higher" functions occur
in the cortex, and since consciousness is a "higher" function,
consciousness must occur exclusively in the cortex. I don't know of any
reason to believe that, just now. The detail with which we can experience
such low-level attributes as brightness, loudness, pressure, acidity,
acridity, and so on would seem to suggest that signals arising at the very
periphery are directly available to awareness. The implication, of course,
is that awareness can receive information directly from any level in the
hierarchy.

Best,

Bill P.

[From Mike Acree (2001.05.23.1445 PDT)]

Bruce Abbott (2000.05.19.2220 EST)--

Ah, a nice discussion of color vision, including Land's Retinex theory

(and

an explanation of his demo) appears on the web at:

http://webvision.med.utah.edu/Color.html

This is probably as up-to-date a review of current thinking and data in

this

area of research as one could hope for . . .

Another excellent treatment, a little older, is Evan Thompson's Colour
Vision (Routledge, 1995), which also includes a discussion of Land's theory.
Thompson, the son of William Irwin Thompson and coauthor, with Francisco
Varela and Eleanor Rosch, of The Embodied Mind, is sensitive to the
philosophical issues, attempting to accommodate both "the mind-dependence of
the world and the world-dependence of the mind."

Mike

[From Bruce Abbott (2001.05.23.2145 EST)]

Mike Acree (2001.05.23.1445 PDT) --

Another excellent treatment, a little older, is Evan Thompson's Colour
Vision (Routledge, 1995), which also includes a discussion of Land's theory.
Thompson, the son of William Irwin Thompson and coauthor, with Francisco
Varela and Eleanor Rosch, of The Embodied Mind, is sensitive to the
philosophical issues, attempting to accommodate both "the mind-dependence of
the world and the world-dependence of the mind."

Thanks, Mike. From what I've seen so far, it would appear that many details
of anatomy have been filled in, together with some information about
synaptic connections (neurotransmitters, inhibitory vs. excitatory
connections), but that even at the retinal level the functional significance
of many of those cells and their connections remains somewhat speculative.

Best wishes,

Bruce A.