D100 and Rebel WTF??? 6MP vs 3MP Opinions please....

I suppose the slate tiles on the tower bridge roofs are aliasing too? Many
aspects of the SD9 image do show more detail than the D60. Call it aliasing
if you like.

 

If you're referring to the tiles on the centre spire, ISTM that neither
camera resolves them particularly clearly. The SD9 image does have lots of
faint thin horizontal lines on it if you look closely, which I assume is
what you're refering to, but I'm inconvinced those are actually the edges of
tiles - the tiles themselves look to be rather larger slabs, so whilst this
particular example shows *something*, I don't think it shows anything
conclusive. Other areas of the image, such as the gibs on the red cranes,
give a slight edge to the D60.

I have performed the suggested experiment, and posted my results earlier in
the thread. The comparison image (10D) is here:

http://www.narcissus.uklinux.net/6vs3.jpg

The image was imported in Adobe Camera RAW iwth no sharpening. The bottom
one was resized to 3 megapixels, and then back to 6 megapixels with bicubic
interpolation. Both then had the same level of sharpening applied.

Even without blowing the image up to examine the individual pixels, it's
brutally apparent that the original premise of the thread does not hold for
this image. The 3 megapixel image looks quiute a lot softer than the 6
megapixel image.

 

It's not that it's "less than perfect", it's just plain wrong. The fence
almost certainly doesn't look like that in real life.

Well, I do find the irregular gaps and thick lines to be less visually
pleasing than the softened haze of the D60 image. Bear in mind that the same
aliasing process that produced these irrecular lines would, when
photographing something with regular horizontal *and* vertical detail,
produce moire. Few people find that they prefer images with moire patterns
in to similar ones without.

Anyway, this comparsion does not support the premise of the thread - the
differences we are seeing here are attributable to aliasing and different
amounts of sharpening between the two images. I performed the experiment
outlined at the start of the thread, and instead of finding that there was
no significant visual difference, as claimed, I discovered that the resized
image ran out of detail significantly before the unresized image. Judge for
yourself:

http://www.narcissus.uklinux.net/6vs3.jpg

 

"The Dude" <.> wrote in message
SNIP

6MP. > I then zoomed into the top of the left tower on each image
(@300%). No

Perhaps if you use a better focused version from a D60, you can?
Try this one (4626 KB):
<http://www.dpreview.com/reviews/CanonEOS10D/Samples/Compared/D60/Outd
oor/d60_CRW_9107_RJ.JPG>

This shows that focus errors can make these comparisons a bit awkward.
Also notice the (slight) difference in magnification between the Sigma
and Canon images. Also look for stair stepped diagonals.

Bart

 

You're comparing 3 MP from a camera with no anti-aliasing filter to 3
MP from a camera that does have an AA filter. That's the main
difference between the two images. This simply doesn't show the effect
that you claim it does (the difference between 3 MP and 6 MP output
from the same camera.)

Dave

 

Yes, honestly. I'd buy the camera on the right even if it cost more.

Consider that when you print the whole frame at any reasonable
size, you are going to be able to see little or no detail in that area
anyway. While if you *do* care about resolving the detail in the
riverboat, you'll use a longer lens. So the "blurry" 6 MP image isn't
really unsharp. (My highest res camera is a 4 MP Bayer type; I'd love
to have the 6 MP 300D or 10D instead).

Basically, the Canon image either shows what was really there, or it
shows a blur. The SD9 image always shows something that looks sharp,
but may not be quite what was really there. I couldn't trust it to do
something predictably faithful to the original scene, even if what it
produces often looks quite good. And a lot of the difference between
the two is in sharpening, which I'd rather do myself anyway.

Dave

 

Better. The tiles are indeed equally resolved in this image.

Magnification? Well, both were shot through Sigma 50 mm F2.8 EX. Jufdging
by the parrallax of the railing, the vantage point is clearly a bit shifted,
but not significantly with respect to the bridge. Distance looks to be
about identical. The other link you sent makes the comparison closer
indeed. Still, pretty damn close for 3MB vs 6MB, no?

Yes. The Foveon in the Sigma has clearly lost some of the highlights where
the Canon hasn't, too -- despite resolving less shadow detail.

 

Agreed. But that does include the well documented advantages of RAW image
processing. My discussion has only included the, admittedly, compromised
jpeg output of the cameras. Can you RAW process to a 3MP tiff? That might
be closer.

 

Don't overlook that the sensors have a physically different size.
Calculated from the sensel pitch, the Canon D60/10D sensor measures
22.6568x15.113 mm, and
the Sigma SD-9/-10 sensor measures 20.68416x13.78944 mm.
So, even with equal Field of View, the Canon sensor is more than 9.5%
larger in both dimensions and thus requires less enlargement for equal
sized output.

Yes, pretty good. However, the lack of AA filter and small fill factor
of the SD-9's sensor unfortunately introduces more artifacts than I
can tolerate (YMMV). Out-of-focus areas e.g. look very strange, just
look at some of the Sigma close-ups others have provided. Single hairs
show as two-three parallel ones, very distracting.

Bart

 

Yes, although I'm pretty sure that all it (Adobe Camera RAW) does is produce
the 6 megapixel image and then interpolate it down, so it's unclear why
there'd be any difference.

 

No. I'm comparing 3MP without AA to a *6MP* with AA. As for the
difference from the same camera, try it yourself and decide. I did.

 

You're right. That was a typo. But the point remains: you're mainly
looking at the effect of removing the AA filter, not of 3 vs 6 MP.

Dave

 

All else being equal, the linear increase in resolution from 3 megapixels to
6 megapixels is 40%. The difference will be a bit less in this case, because
the SD9 lacks an antialias filter, which will give a slight resolution
increase at the expense of artifacting. IOW, we're looking at a difference
is actual resolving power of somewhere in the region of (sticks finger in
air) 30ish percent. It's not that surprising that the images look similar.

Maintaining detail and colour fidelity as the image approaches blowout is
one area where the Canon sensors excel, and the Foveon sensor needs some
work.

 

SNIP

Imagine what would happen if the next Canon DSLR, say an 8MP EOS-3D
(!) had a user removable AA filter...
Besides a dust and aliasing issue, it would kill Foveon, and spark a
whole new discussion ... ;-)

Bart

 

Heheh... careful with the "kill Foveon" stuff. Might get nippy in here

 

So who wants to compare two cameras that produce bad images? It would
not be an improvement if you removed the AA filter from the Bayer
sensor!

Dave

 

Hello
If you go to Luminous-Landscape.com, there are reviews of digital backs
for medium format cameras. The USP seems to be that these backs do not
have AA filters. Admittedly these backs have very resolution, but the
pixel density must be similar to smaller sensors.

The reviews and the results are excellent.

Mike Engles

 

That's a somewhat different situation, for two reasons.

First, medium format lenses tend to have lower resolution per mm than 35
mm camera lenses. The format is much larger, so the total usable
resolution in line pairs per picture height is larger for medium format,
but the line pairs per mm is lower. So the lens blur spot is larger for
medium format, and if the blur spot gets sufficiently large (1-2 pixels
in diameter) it provides all the low-pass filtering needed. (This may be
Kodak's rationale for leaving the filter out of the Kodak 14n, too,
since its pixel density is high).

Second, some of these backs are scanning backs that acquire 3 colour
samples per pixel (using well-separated colour responses, so this isn't
equivalent to Foveon sensors, in case George is reading). Even if such
sensors suffer some spatial aliasing, they don't have any luma/colour
crosstalk problems, so the images may be more tolerable even with some
aliasing.

A Bayer sensor suffers misinterpretation of high-frequency monochrome
detail as colour, if the detail is the right frequency. Thus, the
antialiasing filter is particularly important for a Bayer sensor.

Dave

 

OMG. Compare the clarity of reflection's in the cat's eyes to...
http://www.pbase.com/image/24660210/original

 

Hello

Certainly in a different class.
Superb! These Sigma cameras, do have their problems, but they are also
capable of stunning images. Sigma must perservere and correct the
problems, if they are correctable.

Mike Engles