At 2:52 PM +0000 12/24/02, olympus-digest wrote:
>Date: Tue, 24 Dec 2002 00:47:53 -0800
>From: Moose <olymoose@xxxxxxxxxxxxx>
>Subject: Re: [OM] Re: Dipping our Toe Into Digital
>
>Joe Gwinn wrote:
>
> >Comments interspersed below.
> >
>Likewise
>
> >>Whoa! Check your assumptions. The Mpix numbers advertised for DCs are
> >>approximately the number of 3 color pixels delivered in the output
> >>(after all that complicated stuff you did). My 2.1 Mp camera produces
> >>1600x1200 pixel images with 19,200,000 individual 3 channel pixels, so
> >>it's really a 1.92Mp camera.
> >>
> >
> >Huh? Check decimal point. Ah: 1600x1200= 1,920,000 pixels, but each pixel
> >is of just one color, red, green, or blue, so we don't have the equal of a
> >camera with 1.93 million tri-color pixels. The ratio of colors is 1:2:1 for
> >R:G:B, so there are 1.92/2= 960000 green pixels, 1.92/4= 480000 blue pixels,
> >and 480000 red pixels. The resolution is set by the green pixels; the other
> >colors are interpolated to fill in the missing values. So, a 2.1 Mpix
> >camera actually has 0.96 Mpix of true tri-color resolution.
> >
>OK, so fool that I am, I assume when people argue/discuss the number of
>Megapixels needed to match 35mm information content that they are
>talking about the 3 color pixels in the output produced from processing
>the raw signal from the imaging device, not the number of individual,
>filtered, image sensing points on the CCD/CMOS/?? sensing device. When I
>see camera specs, the number of 3 color output pixels is roughly equal
>to the advertised Mp of the device. So you are talking about sensor
>design detail and I'm talking about device output, however produced by
>interpolation, mixing, mashing or mulching, no? I don't particularly
>care about what's underneath, as I'm empirically estimating the number
>of advertising spec. Mps needed to produce what I consider a great print
>of a large size from the number needed for a smaller print. I realize
>what's underneath counts, but I read the reviews and will only consider
>a camera with good review results.
Not quite. The problem is "marketing pixels". What I call a "tricolor pixel"
has three intensity values, one per color. By contrast, a marketing pixel has
just one intensity value, allowing only one color per such pixel. The most
honest way to state camera capability would be to quote only the count of
tricolor pixels, but the marketing intent was to inflate the perceived
capability of the camera, so each color is counted separately.
How does one translate between tricolor pixel counts and marketing pixel
counts? It depends on the camera design. In the high-end studio digital
cameras, red, green, and blue pixels are present in equal amounts -- the ratio
is 1:1:1. In the consumer and posumer cameras, there are twice as many green
pixels as there are red or blue pixels -- the ratio is 1:2:1.
Lets take a camera with 2.4 Mpix (marketing pixels):
If the ratio is 1:1:1, there are 0.8 million red pixels, 0.8 million green
pixels, and 0.8 million blue pixels. The spatial resolution is 0.8 million
pixels, with full color fidelity.
If the ratio is 1:2:1, there are 0.6 million red pixels, 1.2 million green
pixels, and 0.6 million blue pixels. The spatial resolution is 1.2 million
pixels, bought at some expense in color fidelity.
This increase in spatial resolution, by 1.2/0.8= 1.5, or 50%, is why the 1:2:1
pattern is used. It works because the human eye is less sensitive to blur in
red and blue than in green, the color where visual acuity peaks.
The cost of a CCD chip is basically driven by its total pixel count, and a 50
0ain in perceived resolution is nothing to sneeze at.
>Interestingly, the same simplistic area math says a 3.7Mp output for
>11x14 is equivalent to 1.9 Mp for 8x10 and that's the size Mike M. says
>gives such good results from the 3.7 Mp (effective) E-10. Steven S.
>touts even larger excellent results, but with the aid of special
>upscaling software.
>
>So which number are the folks who come up with the 18Mp as the point
>where digital may match 35mm using?
They are doing some calculation like the one I did a few postings ago. One
assumes a resolution for the film. I chose 50 line pairs (=100 pixels) per
millimeter. Then, one computes the numer of resolution cells in the image in
question. Each cell contains all three colors, so a cell is a tricolor pixel.
Then, convert to markeing pixel counts.
> >Actually, one needs always to look for the "optical resolution", as the
> >1600x1200 may have been interpolated from a lesser number of actual CCD
> >pixels (of any color). The true resolution cannot exceed the optical
> >resolution, regardless of the nominal resolution of the format.
> >Interpolation cannot supply the missing picture detail.
> >
> >
> >>So, using your approach, but adjusted for the way Megapixels are quoted
> >>for DCs and adjusting for promotional inflation, you need about 9 Mp to
> >>equal 35mm film (even less for 25mm film!) I'm not necessarily agreeing
> >>with the 9 MP number here, just disagreeing with your assumptions and 26
> >>Mp conclusion.
> >>
> >Not so. My claim is that one needs more like 25 Mpix (sum of red, green,
> >and blue pixels in 1:1:1 ratio), which is the equivalent of 8.64, call it 9
> >million tri-color pixels.
> >
>I thought that's what I said too, it's just that we're talking different
>ways about the numbers. So what is the 14Mp Kodak? I'm guessing that by
>your calculations, and assuming 1:1:1 sensor design, that you would call
>it a 42Mp sensor?
No, the 14 Mpix is probably the marketing pixel count of a 1:2:1 camera, so the
true spatial resolution will be about 7 Mpix (green only).
As a rule, perhaps except for scientific applications, published pixel counts
are marketing, not tricolor.
> >>Where it comes to where the rubber meets the road for me, all these
> >>calculations don't mean much, it's the images and how people
> >>react/interact, "see" them. It's clear to me that digital camera output
> >>has certain qualities that differ subjectively from film and scanned
> >>film. In the particular case of my eyes and those of friends and family, DC
> >>prints are superior to 2720 dpi scanned 35mm prints for certain
> >>common subjects at 8x10 and smaller. Assuming that's about the limit for
> >>1.9 Mp, one would need about an advertised 8.4 Mp for 16x20, which is
> >>about the limit for sharp 35mm prints using lenses of the quality you
> >>assume and reasonable technique.
> >>
> >>
> >
> >While beauty is in the eye of the beholder, I would comment that there is a
> >lot of equipment in the chain from scene to photograph in each case, and if
> >any element in the chain isn't up to standard, the photos won't be good.
> >So, I would be cautious about making sweeping generalizations from two
> >photos or two specific sets of equipment.
> >
>I generally agree, although less for the DC than for the Olys. Most of
>the S110 images are ready for printing, except sometimes for cropping,
>right out of the camera. Reviews showed that the diminutive S110 had
>some compromises in performance compared to its physically larger
>siblings, so I don't think I am using some unusually great 2 Mp DC.
There seems to be general agreement that 2 Mpix is enough for snapshots, if the
camera is well-designed. People who already have and use computers are often
drawn to the convenience of a digital snapshot camera, although they are often
put off by the amount of equipment and fiddling needed to achieve good results.
And knowledge: the photos from my company's XMAS party are terrible, not so
much because a 2-Mpix camera was used, but because the people doing the
printing don't understand how to use the tools, and are forever exceeding the
gamut of the inkjet printer. The photos may also be overexposed, but I would
start with the printing. Anyway, peoples' faces are blown out, so the photos
look two dimensionsal, with cardboard people. etc. Much worse looking than one
would see with a $10 disposable camera and lousy photo processing.
< This message was delivered via the Olympus Mailing List >
< For questions, mailto:owner-olympus@xxxxxxxxxxxxxxx >
< Web Page: http://Zuiko.sls.bc.ca/swright/olympuslist.html >
|