objects and you can't know that from the spectra or
the spectral distribution energy in the stimulus.
Why? Because the spectral contribution of
the illumination reflectants, transmits and many other factors that lead to
the spectra that make up the stimulus, they're all entangled in the stimulus and
you have no way of getting back to the contributions
of the physical properties of the real world that contributed to the stimulus
that in some sense you need to know about to act appropriately.
So again, taking the same tact that we took in trying to explain
the phenomenology, the discrepancy of our perceptions of black and white
from the actual luminance values that are coming to retina in an empirical way.
You remember we took a database of natural scenes and
we used that database to ask, What's human experience always been?
And are these discrepancies in black and white, and
now we're talking about discrepancies in color.
Are they due to the experience that we've always had,
that's been used by evolving visual systems over evolutionary time,
to resolve this inverse problem to get around the fact that we don't really
have the information about the physical properties of the world in retinal images?
It's just, for the reasons that I've said, impossible.
We don't have that information so we need somehow to get around this problem.
And in color the argument is that, well, okay, we could do it in just the same way,
at least in principle.
That is, we take a whole variety of scenes,
now spectral scenes, that are colored scenes like this and
ask, what's our experience again under the assumption
that over the millions of years, the few million years of
human evolution the natural world hasn't really changed,
what is the experience that we've always had in terms of color.
And again, let me remind you that we can't take these scenes entire
because we never will have seen this exact scene in the same way, more than once.
I think realistically it's fair to say that we never see it more than once but
at least very, very rarely.
So if you want to learn something, whether it's over the course of evolution or
abetted by your experience in your individual lifetime,
you need to see, or you need to take advantage of, the color relationships,
the spectral relationships that exist in little patches.
In the same way we took little patches, more or
less the size of receptive fields of retinal ganglion cells,
of cells in the lateral nucleus in the primary visual cortex,
we need to take little patches because those are the things we see again and
again and again and can learn from.
We can learn what the color relationships are by learning those relationships,
again, not learning them in the sense that we have any knowledge of
this as we normally think about it, but learning them in the sense of having
evolution-generated characteristics in the physiology and
anatomy of our visual systems based on this.
That we can take these little patches, and by analyzing millions of them,
see what the relationships are of the spectra that we have seen forever or
at least forever, forever being the time of human evolution.
And by doing that, by taking millions of these little patches and
asking what are the relationships that we've seen over and over again,
what are the spectral relationships that our visual systems have used to get around
the inverse problem by sampling color experience?
Again, in a database we can understand what our perceptual experience has been
and explain, in the same way as explaining the phenomenology of black and
white, we can explain the phenomenology of color contrast and color constancy
in exactly the same way that I went through and described to you before for
spectra just as for samples from a database of black and white images.
So this is really the same general approach
to solving the same problem in two different contexts.
One, the context of black and white, which is just the overall intensity
of the experience that we've had forever of the patches in natural scenes versus
the spectral relationships in patches from natural scene, such as the ones
that you see here, just as an example of applying the same approach
to the same problem, in the context of color, instead of black and white.
