The Human Eye: 120 Megapixel Monochrome, 6 Megapixel Color

look at it again.

 

So you admit that you're a troll?

 

Now you are being stubborn.

Having three colors at each sensor location is not efficient
if you want to take photos aimed at humans too look at.

It is more efficient to e.g.

1. misaligen the color sensors to achieve more resolution
with the same amount of sensors.
2. have more B&W sensors and fewer color sensors for
exactly the same reason
3. have sensors of different sizes to increase dynamic range.
4. etc

Now - do you still think that haveing three color sensors in
the smae location is the most efficient, measured agians number
of sensors?


/Roland

 

In message <>,

With any given budget of photosensors, this is probably true; with any
given budget of pixels, it is false, all other things being equal.
--

 

Don't be silly. The only part of a Bayer sensor that matches human vision is
the use of RGB.

Bayer pattern sensor are at the moment the best technical compromise for
making silicon based sensors for still-picture photography.

Anyhow, I don't think that the limits of human vision can be translated to
image resolution. I want as much resolution as I can because I really like
to walk towards a poster-sized print and basically zoom in on the image.
It is the difference between looking at a picture and the pervasiveness of
'virtual' reality.

 

It isn't RGB in the eye. The eye pigment sensitivities are much closer
to YGB. R is generated by interpretation in the brain as Y-G. You can
have some fun with perceived colour using a filter that is a narrowband
notch reject against yellow light.

But colour detail is subsampled in the eye relative to the intensity
data. And it is a compromise adopted for broadcast colour TV too - and
for good reason the human eye cannot see colour detail so fine as
intensity.

Full field of view to 1' arc resolution at closest focus gives an upper
bound.

Regards,

 

I know that yellow is special in some sense. I think it has something to
do with the fact that both the 'red' and the 'green' cones are quite
sensitive to yellow, but 'blue' ones are not.

The old analog component video (Y, B-Y, R-Y) is not subsampled. Three
chip video cameras subsample when they encode the data (composite, Y/C,
or any of the digital formats). In retrospect, I think that broadcast
colour TV is a very bad backward compatibility hack.

For obvious reasons, film does not subsample color information.

The problem with subsampled color data is that you have to make sure that
you don't enlarge beyond a certain size. A soft image still looks kind
of natural, but when the colors are wrong, it looks weird.

Both broadcast colour TV and Bayer pattern sensor generate weird colors
as a result of aliasing errors.

It doesn't. That assumes a fixed viewing distance. When you can very
the distance, or if you can walk around, there is almost no limit to
the number of pixels you can use.

Unless you mean, the field of view at the largest distance, together
with the required resolution at the closest distance.

 

Shill.

 

Human vision doesn't use RGB. It uses yellow/green, green and blue.

 

We are debating semantics. Of course if you have six individual colour
million sensors it's better to deploy them in a 6MP Bayer pattern rather
than having a two million full colour pixel CCD.

But six million full colour pixels are of course better than six million
single colour pixels, which is why the Foveon technology is a step
forward.

All we need now are higher MP Foveon CCDs (with more than just 3.4 MP)
and better performance. It's entirely possible that all CCDs will be
Foveon one day.

 

It is not semantics. Three sensors in the same location is
probably as expensive as three sensors in different locations.

There are no six million full color Foveon sensors, only 3.4.
And that for a reason. If you try to put 6 million pixels at
the size of a Foveon sensor you will get very low fill factor.

You might be right. It does not look that way yet though.


/Roland

 

(Philip Homburg) wrote in

The eye does not use RGB.


/Roland

 

I guess I should have written something like 'a linear combination of the CIE
tristimulous respons curves'.

 

Perhaps. It's also possible however that the individual colour sensing
layers in a foveon-type sensor will never be as good as the individual
colour sensing pixels in a bayer-type array, owing to them being
necessarily more complex devices. In which case the bayer compromise
may be better than the foveon compromise.

Or we may end up with a different system entirely. You could for
example use a half-silvered mirror to send most of the incoming
light to a high-resolution monochrome sensor, and the remainder
to a somewhat lower resolution bayer sensor, and combine the
results in firmware. This would also allow for excellent black-and-
white performance. Another possibility that is the subject of a
recent PhD thesis is the use of a two-layer sensor combined with
a two-colour filter pattern.

The problem with all these alternatives is that the bayer compromise
is hardly a compromise at all. Reduced chroma resolution is
completely irrelevant in a photographic context, and the existing
implementations have coloured moire artifacting well enough
under control that this could prove to be similarly unimportant.

- Len

 

Not to mention that as sensors get larger, any artifacts or quality
issues due to anti-aliasing will be less apparent in the final result.
Since making a single-layer Bayer sensor will probably always be
considerably cheaper per pixel than a multi-layered sensor, it's hard to
see a future with multi-layer sensors in it.

 

In fact the eye works just like a Foveon sensor, which is exactly why
the Fovea was named after the Foveon sensor. Seriously, the fact that
the Foveon sensor mimics the eye exactly is why it was named after the
sharpest full color portion of the human eye.

 

Wait, you're saying the FOVEA was named after the FOVEON sensor? You're
hilarious!

 

SNIP

Wrong, again, as usual. The cones in the retina that enable color vision,
don't sense full color each, they sense a partial spectrum.

Bart

 

What a bullshit, the cones and rods are placed besides each other, not
in layers...

Waldo

 

Thius caught me by surprise, so I looked it up (something you might
want to try).
You're wrong.

" the Fovea was named after the Foveon sensor."
Ah, I get it now.

Bill Funk
Change "g" to "a"