18 megapixels on a 1.6x crop camera - Has Canon gone too far?

John A. wrote:
On Mon, 07 Sep 2009 08:22:35 -0500, "mcdonaldREMOVE TO ACTUALLY REACH
wrote:

As far as round (circular) sensors, one answer there is obvious,
in addition to teh fact that they are traditional: You can't
tile a plane with circles ... of useful shapes, only rectangles
and hexagons need apply.

I believe he's talking about covering a round area with photosensors,
not covering an area with round photosensors.


That's what I was referring to. Draw a bunch of circles, say
40 mm in diameter, on an 8 inch diameter silicon wafer. Fill up ...
as close to the edge as possible ... those 40mm circles with
little bitty photosensors, each say 5 microns across,
in a standard Bayer array.

The problem is, you are thus wasting expensive Si area
between the circles. With standard rectangles, there
is no such waste.

Doug McDonald

David Kilpatrick wrote:
Alan Browne wrote:

Excellent description. This is like a radar with a 1.5 deg beam width
detecting a 10 cm diameter pole half a km away when the beam is 13
meters wide by that point. The pole appears to be nearly 26 m wide on
the polar plot as it "paints" from the leading edge of the beam until
the trailing edge... (unless specific "beam sharpening" algorithms are
used which sharpens the plot). But it certainly is detected (perceived).

Alan, you've maybe provided the answer - in a different way.

The human eye doesn't have either sensor or lens image stabilisation!

I know my left eye flickers a bit now as I get older. Sometimes I can
notice it when tired. But our whole body/head as well as eyes are in
constant motion.

So, it's a bit like the sweep of the radar beam - the image on the
retina will NEVER be falling in a static position on those cones. It
will be travelling across/around just like the image you see through
live view camera with an unstabilised tele lens if you magnify it to focus.

We don't see that effect unless there is something wrong (drunk, ear
problems etc) but even when we look closely at one point our eyes are
dancing around it. Plus, there's a pair of them, effectively increasing
the theoretical resolution worked out from the cone density.

Vision is effectively not focusing a static image, like a camera on a
tripod. It is constantly scanning and rescanning across the detail,
measuring depth with stereoscopic vision. We appear to see a static
image but on the retina it is anything but static.

Well, on top of all that, there is also what I mentioned earlier. You
might recognize a particular guitar string at a distance for other
reasons like its curl, its colour/contrast, etc.

IOW, where an "image" has particular information, our brains interpret
context to fill in information about what we think we see.

Alan Browne wrote:

IOW, where an "image" has particular information, our brains interpret
context to fill in information about what we think we see.

Prof Harald Mante did some excellent demonstrations in the 1970s to show
that if you took a photo with small parts of items included in the
frame, but most of the objects cut off, students could complete the
photo by drawing the completed items outside the shot. He thought the
brain also completed these objects/subjects, giving things 'intruding'
into a photo more significance than just the bit shown.

This puts a whole new slant on the 1.6X crop factor - just let the
imagination fill in the missing full frame!

David

David Kilpatrick wrote:
Alan Browne wrote:

IOW, where an "image" has particular information, our brains interpret
context to fill in information about what we think we see.

Prof Harald Mante did some excellent demonstrations in the 1970s to show
that if you took a photo with small parts of items included in the
frame, but most of the objects cut off, students could complete the
photo by drawing the completed items outside the shot. He thought the
brain also completed these objects/subjects, giving things 'intruding'
into a photo more significance than just the bit shown.

There used to be a contest in a local paper where a small part of a well
known personality was shown and you had to indicate who it was. Even
though I had only seen a particular French news anchor a couple times,
and usually from the front, seeing a close crop of the corner of his
mouth from the side was unmistakable.

This puts a whole new slant on the 1.6X crop factor - just let the
imagination fill in the missing full frame!

eh ... no thanks.

David Kilpatrick wrote:

Vision is effectively not focusing a static image, like a camera on a
tripod. It is constantly scanning and rescanning across the detail,

And it must. If you were to project a static image at the retina,
you'd see ... nothing. The retina adjusts permanently and locally
to the input. The brain then consolidates the images (see also
saccadic eye movement).

measuring depth with stereoscopic vision.

Usually vastly overrated, otherwise the illusion of depth wouldn't
work with 2d films.

We appear to see a static
image but on the retina it is anything but static.

If you *test* what resolution one can observe, this doesn't matter.

-Wolfgang

["Followup-To:" header set to rec.photo.digital.]
John A wrote:

It has been noted elsewhere in the thread that people can detect
information from an image beyond that they can directly resolve.

So can a camera. Put a black hair on a white background and
choose a distance and focal length so that on the average 1/4th
pixel is occupied by the hair.
Will you see it? Of course, the pixel only gets 75% of the light.
Will you resolve it? No chance.
Can you get the hair's thickness? Yes, you can estimate it
on the basis of how black the hair-pixels appear. The
more sensitive you are to luminance, the better you can
estimate the thickness, based on your knowledge how
reflective black hair is. You might be fooled by a
colourless grayish hair that's thicker, though.

My point is that perhaps the human eye can gather some information
about an image beyond that in can physically resolve in a photo as
well.

Sure can, but can it do any better than in a photo that
doesn't exceed the resolution of the eye and thus contains
the same information predigested? I don't think so.

Thus printing detail smaller than the eye can resolve may in
fact result in a better-looking, more subjectively natural-looking,
image.

From "perhaps" to "thus" in one sentence. Perhaps you are mad,
thus you should be locked away ... no judge would agree to that
reasoning. Feel free to make the test. Print 2 images, one at
high resolution and one *properly* downsampled at lower resolution,
then blind-test by having random people look at them from larger
distances so that they absolutely cannot resolve the first image.
If your theory is true, ... Obvious, isn't it?

Or perhaps not. But it does seem like the human visual system is
capable of more than just directly resolving edges and widths.

Nobody disagreed, last time I looked.

And then, of course, there's the whole cropping and poster print thing
that always seems to get lost in these debates. People do more with
the pictures they take than just print the whole thing at 8x10.

Posters get printed at, what, 150dpi at most? Pardon me, but
isn't that a bit low? Or are they just to be looked at from
a distance?

-Wolfgang

["Followup-To:" header set to rec.photo.digital.]
"mcdonaldREMOVE TO ACTUALLY REACH wrote:
I have an 8 MP Canon 30D. I can tell the difference between
the sharpness and general quality of a 4x6 print made from it,
using a wide angle lens (24 mm), and a print (same size) made
from a panorama assembly of images made with a 50 mm lens,
i.e. roughly 20 (one loses some due to overlap). In both cases
the lens was used at f/9, its optimal aperture, and lateral
chromatic aberration correction was used in Canon DPP,
and identical amounts of sharpening.

I see. So you can tell the difference between an 8 MPix and an
160 MPix image on a 3.8 MPix print (4x6 inch at 400(!) ppi).
Or are you telling us you have a 2582 ppi (not dpi) printer
stashed away somewhere?

What you are seeing is bad downsampling (or vignetting or
somesuch).

And what is "identical amounts of sharpening" when the same
radius of x pixels has a completely different impact (and
thus sharpening effect) on the 3.8 MPix resulting image?

-Wolfgang

"mcdonaldREMOVE TO ACTUALLY REACH wrote:

As far as round (circular) sensors, one answer there is obvious,
in addition to teh fact that they are traditional: You can't
tile a plane with circles ... of useful shapes, only rectangles
and hexagons need apply.

You can tile a plane with circles, perfectly even, given that
wafers are round, in the special case of N=1. Now all the old
wafer systems with only 20cm or less in diameter get a new life,
and we get round sensors of 15 or 20cm diameter. And since we
don't really need the ultra-dense circuity in first place (nice
as it may be --- we are going for *large* photosites this time
around), we can reuse all the old, cast-off gear no longer used
by computer chip makers.

Now we just need some larger bodies and lenses with, say, 20+cm
image circles. Designing the mirrors and shutters for DSLRs may
be fun, too --- I suggest 2 counterrotating disks as mirror to
eliminate mirror slap and vibrations --- huge, but workable ---
and possibly the same design for the shutter. Or maybe a
central shutter after all.

And Oly can take away the mirror and use EVIL models with much
thinner bodies, to avoid the usual wide angle retro focus
design problem.

Now we only must hire someone to carry the new cameras and lenses
for us and we all shall be happy.


Alternatively, one could accept some wastage when tiling up
circular sensors, attach camera CPUs directly to the sensor in
the 'wasted' space and fill out the rest as well as possible with
smaller, different chips. It's not like camera chips are going
to shrink anyway.

-Wolfgang

Elliott Roper wrote:

I'm arguing that "35mm" is a backward looking nomenclature for digital
camera sensors.

So what? Your name is a backward looking nomenclature for you,
unless it's gallows bird or jail bird ...

I'm arguing it is made worse by describing cameras as fractions of 35mm.

Why would something that's factual, but not bad be made worse
by something that's factual and not bad, either?

I'd like a straightforward naming scheme based on sensor dimensions.

Fractions of 35mm frames isn't bad, then.

Starting from an obsolete recording medium is helpful only to those
with experience of those old systems.

Since there are exactly *absolutely ZERO* people with experience
in your new system, changing the system would cause harm without
any advantage to counter it at all.

Asserting that you own a large number of lenses suitable for that
obsolete system is irrelevant.

Asserting that your new system is in any way better is vastly
more irrelevant, since the buying power in that assertion is
zero, whereas lenses will be a solid investment.

If anything, the assertion shows you are uncomfortable with thinking
about sensor sizes in a rational way.

Rational? Like ... fractions?

Oh wait! You *do* understand. Later you call them 44mm lenses. That is
a better name than 35mm.

Better in what regard? Some artificial measurement you want
to push?

It means they create a 44mm diameter focused
image. So why persist in calling the lenses 35mm? Is it a comfort
blanket?

Why do you persist in pushing new names *noone* can relate
to? Do you want everyone to feel like you do, left out?

A 44mm lens is just the right size for a 24*36mm sensor. That is
obvious to anyone with a calculator and a bit of elementary maths.

Ah, so you do lens choosing with calculators and math, but
are completely unable to handle fractions? Dear me!

It is also obvious that it would be OK for a smaller sensor.

Try adapting it to your usual point'n'shoot and show us how
OK that is.

With that naming convention anyone can do the maths to see how wasteful
and what angle of view.

But you cannot do fractions?

But where does 35mm fit? In the past.

My, are you ever so bitter.

Using the camcorder sensor size naming mess is completely relevant to
this discussion. They too fell into the trap of describing their
sensors relative to obsolete hardware.

Well, you would describe sensors on thin air instead. Better to
offend everybody than to offend only some, a very good plan indeed.

Asserting that there are a lot
of 44mm lenses out there is no refutation of that relevance.

The relevance is that the 35mm sensor is going to stay around
for a *long*, *long* time.

Finally, who is this "we" that does not care about small sensors in
fixed lens cameras?

Most everybody but you.

A decent sensor and lens naming scheme should take
account of them.

Yes, please make a new naming scheme for all the focal lengths,
the world has waited for that for a long time. Call the lengths
things like "square root of 3/4th froob" and "9/7th quux cubed".
That surely will offend everyone and thus be ideal.

used a proper name like 15*22.5mm you would have less chance of getting
your fractions upside down. And you would know that a 27mm lens would
be adequate.

How would *you*, who isn't even able to deal with fractions,
know about 27mm? Calculator?

Yet you say that old 35mm naming convention isn't broken.

It works pretty well, unlike your scheme. Bet you had to look
up 15x22.5mm.

-Wolfgang

In rec.photo.digital David Kilpatrick wrote:
Alan Browne wrote:

Excellent description. This is like a radar with a 1.5 deg beam width
detecting a 10 cm diameter pole half a km away when the beam is 13
meters wide by that point. The pole appears to be nearly 26 m wide on
the polar plot as it "paints" from the leading edge of the beam until
the trailing edge... (unless specific "beam sharpening" algorithms are
used which sharpens the plot). But it certainly is detected (perceived).

Alan, you've maybe provided the answer - in a different way.

The human eye doesn't have either sensor or lens image stabilisation!

I know my left eye flickers a bit now as I get older. Sometimes I can
notice it when tired. But our whole body/head as well as eyes are in
constant motion.

So, it's a bit like the sweep of the radar beam - the image on the
retina will NEVER be falling in a static position on those cones. It
will be travelling across/around just like the image you see through
live view camera with an unstabilised tele lens if you magnify it to focus.

We don't see that effect unless there is something wrong (drunk, ear
problems etc) but even when we look closely at one point our eyes are
dancing around it. Plus, there's a pair of them, effectively increasing
the theoretical resolution worked out from the cone density.

Vision is effectively not focusing a static image, like a camera on a
tripod. It is constantly scanning and rescanning across the detail,
measuring depth with stereoscopic vision. We appear to see a static
image but on the retina it is anything but static.

Not quite. It's true that the image we "see" is constructed in the
retinal map areas of the brain from many different snapshots taken as
the eye dances around, and includes a lot of pretty heavy (in
photographic terms) post processing to do such things as erase the
constant shadows of the vascular and nervous twiggery that runs over
it, filling in the blind spot, filling in strong expectations based on
slight evidence, etc.. But the "dancing around" is done very jerkily
by very fast saccadic jumps during which the eye is blinded, followed
by stationary intervals during which the image is recorded.

It's also not true that the eye doesn't have image stabilisation. It
has two different kinds of image stabilisation. The first is the
learned stabilisation of keeping the eye focused on the point of
attention despite movement of the head or body. That allows you to
keep your eye on prey you're chasing down, or on the fists and weapons
of an enemy whose blows you're dodging.

The second is distinguishing between movement of your eyes and
movement of the world. That's why the world doesn't seem to move if
you move your head, walk, or move your eyes. but does seem to move if
you poke your eyeball. And is sometimes fooled, as when you think your
train has started moving when it's the other train you're looking at
that is moving.

One of the reasons why so much care has been taken to "snap" the eye's
images when the image is stabilised is that at a very low and very
fast reacting level of visual processing in the retina and early brain
retinal maps it is exquisitely sensitive to small movements anywhere
in the field of view. That's how we spot lurking animals who might be
dinner or who might think we were dinner. That sensitivity to movement
in an otherwise stationary image would be lost if so much care to
stabilise the image wasn't taken.

An interesting demonstration of this is the experiment that can be
done with a computer screen of text. We think we can see all the text,
some of of it well enough focussed to read. But that's an impression
stitched up like a panorama from constant flickering saccadic
snaps. If the screen viewer's head is locked down and the gaze
direction monitored you can write software which scrambles the screen
image text except for the few words the eye is pointing to. The
scrambling is done during the saccadic jumps, and the screen is stable
when the eye is stable and seeing. The viewer thinks he's looking at a
completely clear stable screen of legible text. But nobody else can
read any of the flickering scrambled mess!

--
Chris Malcolm