Re: More pixels are better, much of the time

"Scott W" <biphoto@xxxxxxxxxxx> writes:

But this is an easy test of this to get a good feel of what you would
be looking at in the two cases. Lets take the case of the pixel on the
20D or 350D, this are about 2.5x as larger as the pixels on my Sony
F828, so a sensor the size of the 20D with pixels the size of the F828
would have right around 20MP. If I want to get a quick look at how a
sensor like this would compare to the lager pixels of the 20D I can
take the same photo with both camera using the same focal length for
both. In the sample the 350D images were taken at 28mm and the F828
taken at 27.5, which is about as close as I can get without a lot of
messing around.

I don't think this test tells you much about how well smaller pixels
would perform on large sensors. What you're trying to do is predict how
828-sized pixels would perform on a 20D-sized sensor behind a 20D-sized
lens. But what you did is shoot the 828 tests with the 828 lens, which
is very very different.

More precisely, you took the 828 lens, which is designed to cover an 11
mm image circle (the diagonal of the 828 sensor) and shot it at 27.5 mm
focal length, which is a medium telephoto for that camera (about
equivalent to 110 mm on full-frame 35). A short focal length lens like
this covering a moderate angle like this should resolve 150-200 lp/mm on
the sensor, but it only has to do it for that 11 mm circle (or in
practice only the central portion of it).

Then you used a 35mm camera lens, which was almost certainly designed to
cover a 43 mm image circle, on the 20D. Because the format it is
designed for is 4X larger than the 828 lens, this lens only needs to
resolve 1/4 as many line pairs per mm for the same final image quality.
And even getting that is more difficult because this is a wide angle
lens, not a medium telephoto (WA lenses are difficult to design due to
the steep angles of the corner rays). You shoot an image with this
lens, and crop out 1/4 of the width and height of the image it was
designed to deliver (1/16 of the area), and then enlarge that to the
same size as the full 828 image.

Under these conditions, *of course* the 828 image looks better. It
would look better even if you put a 20 MP sensor in the 20D with the
same pixel pitch as the 828. You're effectively demonstrating that 16 8
MP medium telephoto images stitched will generally look better than one
single 128 MP wide angle image covering the same angle. Of course it
does.

There are two ways to view the comparison, enlarge the 350D image to
match the F828
http://www.pbase.com/konascott/image/66254346/original
Here the 100% crop from the F828 is on the left and the 350 image is on
the right, resized to 250%
Both photos were taken at ISO 100.

No surprise. You're demonstrating that shooting a much smaller area of
the subject with a lens that provides more magnification captures more
detail. But change that 28 mm lens to a 70 mm, to cover the same FOV on
the 20D, and I'll bet that the 20D image will become comparable. Or
better yet put a 110 mm lens on a 5D, and use the full image circle of
the lens while capturing the same FOV.

It also looks like the default sharpening of the 828 is more than the
default sharpening applied by the 20D, and you haven't tried to make
these as comparable as possible.

We can also compare the two by resizing the F828 image to 40% to have
it match the overall size of the 350D image. When we do this we
greatly reduce the noise of the F828 image and still end up with a
sharper image.
http://www.pbase.com/konascott/image/66254320/original

Yeah, but you're still comparing the full useful image from one lens to 1/16
of the image from the other lens.

Now in order to get an image as sharp as the F828 over the full 20MP
sensor you would need a very good lens indeed, but that is another
issue.

I think your demonstration mainly shows the difference in lenses, not in
pixel pitch. So whether you could get a lens good enough is the *central*
issue in whether this many pixels is practical. Even if you designed
a 28 mm lens to cover only the 27 mm diagonal of the 20D, not the 43 mm
full-frame diagonal, achieving the same 150+ lp/mm as the little 828
lens with its 11 mm image circle is going to be very difficult.

Looked at another way, the pixel pitch of current DSLR cameras is
already reasonably matched to the resolution of most of the lenses they
will be used with. Unless you can improve the lens resolution by almost
2.5X, there isn't going to be much point in decreasing the pixel pitch
by 2.5X.

Dave
.

Relevant Pages

  • Re: description of camera design I would like to have
    ... element and the sensor than is current practice with DSLRs. ... The specifics are that when Leica decided ... Leica is the only major manufacturer of interchangeable lens 35mm ... on their specific design. ...
    (rec.photo.digital)
  • Re: description of camera design I would like to have
    ... high end super-pro camera market business decisions ... had nothing to do with making a better lens design. ... plane can have more issues with a digital sensor than with film. ...
    (rec.photo.digital)
  • Re: Film versus digital sensor lenses question
    ... In particular, something in the sensor design, either the relative ... presence of micro lens, meant that the lens had to be designed ... pixels instead of 1 million pixels? ... angles the rays of light from the back of the lens meet the sensor. ...
    (alt.photography)
  • Re: Full-frame or 1.5 DSLR?
    ... >16.7) FF cameras can resolve subject detail as well as the 20D (or ... >But the lens has to deliver its resolution to the sensor, ... >matter that the sensor has more pixels per mm in the 20D sensor, ...
    (rec.photo.digital.slr-systems)
  • Re: Low cost image sensor
    ... >I have been searching the web for a low cost image sensor. ... read to 0.1mm resolution which actually means a minimum of 200 pixels ... realistically 400 pixels would be better to ensure the resolution. ... Ideally I would like no moving parts, and no lens. ...
    (comp.arch.embedded)