Re: Fovean Sensors
- From: davem@xxxxxxxxx (Dave Martindale)
- Date: Fri, 25 Jul 2008 16:48:32 +0000 (UTC)
2SQUID <2squid@xxxxxxxxxxxxxxxxx> writes:
The SD9/10 resolve as well or better than any other 3.5 MP Bayer camera,
but they're not at all comparable to a 10 MP or 14 MP Bayer camera in
real resolution.
Is this from your personal experience Dave? It doesn't sound like it.
I don't own one of the SD cameras. But my opinion is based on examining
a number of full-size images posted by Sigma owners, and the sample
images on the dpreview pages.
I put this page together an hour or so ago:
http://www.auspub.com.au/squids/SD14.htm
There's very little to look at in the 100% scale image, but the fine
hair detail is a mess. Individual strands show bad "staircase"
artifacts at some angles. Now, have you tried comparing this to the
image from a 10 MP Bayer camera resized down to match the Sigma's image
size?
You might not agree with my opinion or that I believe a "real"
photographer needs to separate themselves from everything except the
photography but this quick and dirty exercise proves 3 things to me.
1. The Bayer camp have done a fine job of fooling people for far too
long. Many 10 and 12 megapixel cameras produce absolutely shocking
images that need an on-board computer to patch up their images before
saving them as some engineers idea of a RAW file.
You have said nothing about what you mean by "shocking", so I'll ignore
that as content-free. Similarly, both Bayer and Sigma cameras store
"some engineers idea of a RAW file", so that's content-free too. In
fact, you don't describe what "patch up" means, so the entire paragraph
is pointless.
2. The Sigma SD14 is every bit the equal to a 10 or 12 megapixel Bayer
image in sharpness and clarity when de-mosaiced to the same size using
Sigma's own Photo Pro software.
Equal in what way? The Sigma image may *look* sharp, but examination
of the resolution test images and real photographs show plenty of
artifacts and less resolution for real fine detail. Now, for some
subject matter, the Sigma actually produces better images overall at
the same number of megapixels. But comparing a 3.5 MP Sigma to a 10 or
12 MP Bayer, the Sigma just always loses, unless you really like the
sort of image mistakes it makes. (And you can get most of those by
downsampling using nearest neighbor in Photoshop; you don't need a Sigma
camera for that effect).
2. That unlike the Bayer cameras, the Sigma does very little "in-camera"
processing and relies substantially on a 'real' computer running Sigma's
own Photo Pro software to finalise the image.
If you want, you can have Bayer cameras write raw data and do the
processing on a host computer too. It does give more control. The
difference is that the Bayer cameras can *also* produce decent JPEGs, or
both RAW and JPEG at the same time if that works better for you. I
don't know if this is because Sigma can't afford a decent signal
processor for the camera, or their algorithms are just too expensive to
run in any available processor, but this is definitely a place where the
Sigma cameras lose to all of the competition.
I also believe that before any meaningful discussion of the resolution
of different types of sensors can be had, we really do need to recognise
that a Bayer sensor absolutely need three times the number of pick-up
sites, just to capture the same colour information as a Foveon sensor
does. This results in the Bayer camp accusing Foveon of fobbing the figures.
First, the colour response of the Foveon sensors is notoriously bad,
because the three channels are not as spectrally independent of each
other as they ought to be (and not compared to Bayer sensor filters).
So the colour transforms involve a lot of cancellation, which boosts
noise. And the cameras are also notorious for blotchy colour and yellow
casts. So Foveon colour is nothing to aspire to.
Second, even in a theoretically perfect world, the real advantage of
sensing colour at every pixel is that it gives you colour-difference
resolution that matches luma resolution. This is nice if you can get
it, but not really necessary for pictorial photography since the human
eye's resolution for colour-only difference is *one tenth* of its
B&W resolution. Bayer sensors effectively have a colour resolution that
is half their luminance resolution - which is better than it needs to be
to satisfy the eye in prints where the luminance seems acceptably sharp.
So the Foveon's colour resolution advantage isn't as significant as it
might seem.
I also make a guess that most of the knockers of the SD14 have never
held one in their hand, much less spent any time using one. I used a
handed down SD10 for several years while I was doing my studies. I got
pasted by everyone and sundry until they saw the life size portraits it
took and the 36" wide seascapes I made from it's images. You simply
couldn't do this if it were resolving to 4 mega pixels.
Handwave all you want, but the sensor physically has only 2268x1512
light-measuring locations.
Your experience is evidence that you can produce pretty good images of
certain subjects with less than 4 MP - because that's all the camera
does have. On the other hand, the image from a 12 MP Bayer camera
downsampled to 4 MP using a decent algorithm will *also* look pretty
good, without the Sigma's artifacts. And if you leave the Bayer image
at 12 MP, it can look even better.
There seems to be a collision going on between photographers and
technocrats that never existed in the days of film. I can't find any
discussion like this about the resolution of film, the way there is in
how a manufacturer measures their camera's resolution.
As I remember it, even in the days when everyone used film, there were
those who liked films with coarse but very sharp grain, and those who
liked finer but "mushier" grain. And there was some controversy about
how to measure resolution. And video cameras didn't look like film
cameras.
Sigma SD14 cameras are on an equal par with any 10 or 12 megapixel Bayer
or CCD sensor. And I'll add this in parting. Backlit subjects are a
Bayer's worst nightmare and a Foveon's playground for exciting images.
It seems to be a matter of preference. Some people look at Sigma images
and think they're beautiful, other photographers look and say "yuck".
Sometimes even when both are looking at the same image.
But, objectively looking at tests, it's clear that (a) Foveon sensors
are not competitive in resolution due to low pixel count, (b) the lack
of an anti-aliasing filter results in various image artifacts in fine
detail. You may *like* the images anyway, but not everyone does.
Dave
.
- References:
- Fovean Sensors
- From: measekite
- Re: Fovean Sensors
- From: 2SQUID
- Re: Fovean Sensors
- From: Dave Martindale
- Re: Fovean Sensors
- From: 2SQUID
- Fovean Sensors
- Prev by Date: Re: 18m 10mm?
- Next by Date: Re: 18m 10mm?
- Previous by thread: Re: Fovean Sensors
- Next by thread: Re: Fovean Sensors
- Index(es):
Relevant Pages
|
