Re: D30, 300D or SD9?

Discussion in 'Digital Photography' started by David Kilpatrick, Apr 1, 2004.

  1. John Navas wrote:
    >


    >
    > A decent flatbed has much higher native resolution -- for example, the Epson
    > Perfection 3170 Photo (US$170 street) has an optical resolution of 3200 dpi,
    > which translates to over 15 MP when scanning a 35mm frame, or about 4.5x the
    > SD10. With Micro Step Drive technology, the hardware resolution of the 3170
    > goes even higher, up to 3200 x 6400 dpi.
    >
    >
    >>And they size up 2X
    >>as well as a good flatbed scan can do.

    >
    >
    > 8x10 at 300 dpi is 2400 x 3000 (7.2 MP), so the scanner would be
    > *down*sampling while the SD10 is *up*sampling.
    >


    Sorry, I only use digital camera files generally at their native
    resolution resized to 300 dpi, same as a scanner. And I never use
    flatbeds for trannies, and generally I'm talking about simple 300 dpi
    flatbeds, and what you might do with a 6 x 4 print.

    Just different worlds John - I don't know what you do but I publish
    magazines and we don't need anything like 3200 x 6400 dpi or attempting
    to scan slides on flatbeds! Gimme an old 300 dpi flatbed any day and all
    it will ever be asked to do is prints. At 100 per cent size max. For
    slides, use a slide scanner...

    Actually I use an Epson 1660 Photo but the only time it's asked to do
    600 dpi or 1200 dpi is the occasional line art logo, and the only time
    the tranny scanner has ever been used was for a panoramic rollfilm shot.

    Whatever the arguments, a typical half-page illustration out of the SD10
    is actually the best standard of half-page repro I've had since using
    scanning cameras in the studio. I'm not stupid and I do understand the
    physics and maths of aliasing and the problems of translating a contone
    world into a cartesian grid of discrete values, and the dilemma of equal
    response to a partial coverage of a sensor by a stimulus of high value
    versus full coverage of the sensor by a lower level of illumination -
    producing identical final signal levels from very different subject
    targets. Etc etc.

    In fact, you can get tangled up in the semantics as easily as the math.

    I just look at the images I and I know what I see, and the SD10 - which
    I would never *ever* have imagined myself using before I completed tests
    on it - hits the buttons. It works. Despite all the reasons why it can
    not work and must not work and DOES NOT WORK and all...

    And I don't think the developers of cameras are blind, and I do think
    they will be looking hard at why it seems to work when it shouldn't.

    David
    David Kilpatrick, Apr 1, 2004
    #1
    1. Advertising

  2. David Kilpatrick wrote:
    >
    > I just look at the images I and I know what I see, and the SD10 - which
    > I would never *ever* have imagined myself using before I completed tests
    > on it - hits the buttons. It works. Despite all the reasons why it can
    > not work and must not work and DOES NOT WORK and all...
    >
    > And I don't think the developers of cameras are blind, and I do think
    > they will be looking hard at why it seems to work when it shouldn't.


    I agree.
    Actual experience is more important than theory - that's a lesson which
    is probably hard to learn for academics.

    I had a similar experience with my JPEG rescaling algorithms.
    I think they are basically rather simple to understand (preparing a paper
    about this...), but (academic) people just couldn't imagine that it worked.

    I had a discussion in the newsgroup comp.graphics.algorithms
    ("How to downsampling still picture on DCT Domain") where some
    guy suggested to use a 6x6 IDCT for 75% downscale, but "experts"
    responded that it can't actually work because of the 'blockiness'.
    Apparently they never tried it - I tried it and got perfect results
    without any sign of blockiness, so that is the best method to downscale
    a JPEG by 75%. This fact is still hard to understand for academics,
    perhaps it doesn't fit well enough in their known complicated theories.

    Regards
    Guido
    Guido Vollbeding, Apr 1, 2004
    #2
    1. Advertising

  3. David Kilpatrick

    John Navas Guest

    [POSTED TO rec.photo.digital - REPLY ON USENET PLEASE]

    In <c4gqcd$6bd$> on Thu, 1 Apr 2004 10:19:58 +0000
    (UTC), David Kilpatrick <> wrote:

    >John Navas wrote:
    >>
    >> A decent flatbed has much higher native resolution -- for example, the Epson
    >> Perfection 3170 Photo (US$170 street) has an optical resolution of 3200 dpi,
    >> which translates to over 15 MP when scanning a 35mm frame, or about 4.5x the
    >> SD10. With Micro Step Drive technology, the hardware resolution of the 3170
    >> goes even higher, up to 3200 x 6400 dpi.
    >>
    >>>And they size up 2X
    >>>as well as a good flatbed scan can do.

    >>
    >> 8x10 at 300 dpi is 2400 x 3000 (7.2 MP), so the scanner would be
    >> *down*sampling while the SD10 is *up*sampling.

    >
    >Sorry, I only use digital camera files generally at their native
    >resolution resized to 300 dpi, same as a scanner. And I never use
    >flatbeds for trannies, and generally I'm talking about simple 300 dpi
    >flatbeds,


    Simple 300 dpi flatbeds are of course a thing of the past. Even entry level
    scanners now have much higher native resolutions; e.g., Epson Perfection 1670
    PHOTO with an optical resolution of 1600 dpi and a street price of only
    US$100.

    >and what you might do with a 6 x 4 print.


    You scan 6x4 prints at 300 dpi? I generally get better results by scanning at
    higher resolution (600x600 or higher), and then downsampling.

    >Just different worlds John - I don't know what you do but I publish
    >magazines and we don't need anything like 3200 x 6400 dpi or attempting
    >to scan slides on flatbeds! Gimme an old 300 dpi flatbed any day and all
    >it will ever be asked to do is prints. At 100 per cent size max.


    I find current scanners to be quite a bit better than the old 300 dpi
    flatbeds, and downsampling to be helpful.

    >For
    >slides, use a slide scanner...


    Of course, if the cost is justified. Otherwise, the better flatbeds can now a
    pretty decent job.

    >Actually I use an Epson 1660 Photo but the only time it's asked to do
    >600 dpi or 1200 dpi is the occasional line art logo, and the only time
    >the tranny scanner has ever been used was for a panoramic rollfilm shot.


    Ah, but then you probably are scanning at higher resolution and downsampling!
    That's because the 1660 is probably scanning at its native resolution of 1600
    dpi, and downsampling to whatever lower dpi you specify. That's a significant
    advantage of a current high-res scanner over an old native 300 dpi flatbed.

    For printing on a high resolution, high quality printer (e.g., Epson R800,
    Noritsu minilab), have you critically examined the difference between scanning
    at 300 dpi *native* versus scanning at (say) 1200 dpi *native* with quality
    downsampling (in the scanner, scanner driver, and/or with image processing
    software) to 300 dpi before printing? I have, and can often see the
    difference (depending on the type of image).

    I generally scan at about 1200x1200 dpi (depending on the native resolution
    of the scanner) in any event because I usually archive the scan for possible
    future use.

    >Whatever the arguments, a typical half-page illustration out of the SD10
    >is actually the best standard of half-page repro I've had since using
    >scanning cameras in the studio.


    As compared to the Canon 1Ds? Visually better than the Canon 10D? Really?
    What about full-page?

    --
    Best regards,
    John Navas
    [PLEASE NOTE: Ads belong *only* in rec.photo.marketplace.digital, as per
    <http://bobatkins.photo.net/info/charter.htm> <http://rpdfaq.50megs.com/>]
    John Navas, Apr 1, 2004
    #3
  4. writes:

    >I just look at the images I and I know what I see, and the SD10 - which
    >I would never *ever* have imagined myself using before I completed tests
    >on it - hits the buttons. It works. Despite all the reasons why it can
    >not work and must not work and DOES NOT WORK and all...


    It is image content dependent. For some subjects, doing point sampling
    without any anti-alias filtering gives images that look *better* than if
    you'd done the theoretically correct thing. This is particularly true
    if you're looking at still images, since you can't notice that the
    apparent image content changes with tiny amounts of camera motion. But
    for other subjects, the extra (false) detail looks unnatural or just
    plain wrong.

    You can experiment with this easily in Photoshop. Find some
    high-quality high-pixel-count image you like. Reduce it in size by
    some integer factor, to 50% or 33% or 25% or 20%, using nearest
    neighbor resampling. Reduce another copy of the same source image to
    the same size, but using bicubic resampling. Look at both small images
    at 100% on screen. The NN image will almost always look sharper, but
    sometimes it will be less natural than the bicubic-resampled one. And
    which one looks better depends on the original image, and your own
    subjective image judgement. (You can also reduce the apparent
    difference by applying a bit of small-radius unsharp mask to the second
    image after resizing).

    Now, if I'm doing this resizing in Photoshop, I can try both and see
    which method I like. With a digital camera, you only get one output
    image. A camera without an AA filter like the SD9 gives you an image
    like the nearest-neighbor resampled one, while a camera with a good AA
    filter gives you an image much like the bicubic-resampled one. When
    choosing a camera, I want the latter behaviour because it's predictable
    and faithful to the original scene.

    (And, in fact, I almost never do size reduction in Photoshop twice and
    compare the results. I know from experience that I generally hate the
    results of NN resampling, so I just use bicubic and some
    post-sharpening. Or I use a different program with other resampling
    filters.)

    Dave
    Dave Martindale, Apr 1, 2004
    #4
  5. "David Kilpatrick" <> wrote in message
    news:c4gqcd$6bd$...
    SNIP
    > Actually I use an Epson 1660 Photo but the only time it's asked to do
    > 600 dpi or 1200 dpi is the occasional line art logo,


    To reduce aliasing I presume...

    > and the only time
    > the tranny scanner has ever been used was for a panoramic rollfilm shot.


    Bart
    Bart van der Wolf, Apr 1, 2004
    #5
  6. John Navas wrote:

    Re:

    >>Whatever the arguments, a typical half-page illustration out of the SD10
    >>is actually the best standard of half-page repro I've had since using
    >>scanning cameras in the studio.

    >
    >
    > As compared to the Canon 1Ds? Visually better than the Canon 10D? Really?
    > What about full-page?
    >

    I would hesitate to use any 3-5 megapixel digital camera for full page
    routinely, though we generally test 6+ models now with a double-page
    spread repro - or as large as the file goes when NOT resampled, at 300
    dpi. The Sigma's true size file (not the interpolated up 2X version) is
    a really nice clean 1/2 page. The Canon 1DS is considerably larger as a
    clean repro but then it's also six times the price!

    One of the things I have had to do for Minolta is to do a full page
    repro from every camera since the 2330 (2.3 megapixels or something long
    forgotten!). Funny thing is, they have all been pretty good, and the 5
    megapixel models have been indistuinguishable from the average glossy
    catalogue illustration - and the same goes for Canon, Casio, Fuji 6
    megapixel (true) and several other cameras. The only real stinker so far
    has been the Contax TVS Digital, which lacked a file format of
    sufficient quality to make a good full page repro (no raw, no TIFF...)

    The Sigma holds up very well against any other opposition at full page.
    I did the SD9 as a double-page spread using their 120-300mm zoom and a
    very low light shot of a robin on a branch - I really didn't know at the
    time how limited its high ISO performance was, but since I didn't know,
    I used it, and it worked surprisingly well in print. But it was faking
    it a bit as most of the background was a nice differential focus blur.
    The result looked much like a DPS from 200 speed colour slide 35mm would
    look.

    The DPS we did from the Canon D1S could easily have been from 5 x 4. It
    was a very different kind of shot, an architectural picture with water
    and sky (I need somewhere to print the text) taken by a contributor
    working near Nottingham University, which had some good subject matter
    and conditions at the time.

    Comparisons with the D1S are invidious, but oddly enough, comparisons
    with the DCS Pro/n are not. Three weeks of trials with the Pro/n left me
    feeling fairly pleased I have decided to use the Sigma for the next few
    months - and I was desperately hoping the Pro/n would be D1S quality
    with a Nikon mount and an even bigger image. It ain't. It should have
    been. Kodak are not pleased with my initial reactions, or those of other
    photographers, but seem more interested in denying or defending, than
    finding out how to get it right.

    David
    David Kilpatrick, Apr 1, 2004
    #6
  7. David Kilpatrick

    Guest

    In message <>,
    Guido Vollbeding <> wrote:

    >David Kilpatrick wrote:
    >>
    >> I just look at the images I and I know what I see, and the SD10 - which
    >> I would never *ever* have imagined myself using before I completed tests
    >> on it - hits the buttons. It works. Despite all the reasons why it can
    >> not work and must not work and DOES NOT WORK and all...
    >>
    >> And I don't think the developers of cameras are blind, and I do think
    >> they will be looking hard at why it seems to work when it shouldn't.

    >
    >I agree.
    >Actual experience is more important than theory - that's a lesson which
    >is probably hard to learn for academics.


    Academics? Like "2/3 of image information missing"? YOU are the one
    with your head stuck up the ass of academics. All your proofs about how
    the Foveon and its Sigma implementations are allegedly superior to Bayer
    are based on arithmetic.

    The images themselves speak otherwise. Both the SD9 and SD10 have
    problems with colors in gradients, especially shades of blue and skin
    colors. The SD10 says, "I'm made of pixels", and the SD9 *SCREAMS* it.
    They blow out very easily, with bizarre color fringing at the edge of
    the blown out areas.

    [irrelevant JPEG stuff snipped]

    --

    <>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
    John P Sheehy <>
    ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
    , Apr 2, 2004
    #7
  8. Dave Martindale wrote:
    >
    > It is image content dependent. For some subjects, doing point sampling
    > without any anti-alias filtering gives images that look *better* than if
    > you'd done the theoretically correct thing.


    The Foveon sensor does more area-sampling than point-sampling, and that's
    not nearly so bad as you suppose.

    > Now, if I'm doing this resizing in Photoshop, I can try both and see
    > which method I like. With a digital camera, you only get one output
    > image. A camera without an AA filter like the SD9 gives you an image
    > like the nearest-neighbor resampled one,


    Wrong. SD9 and more SD10 do *area*-sampling, not point-sampling.
    Therefore the comparison with nearest-neighbor resampling is wrong.
    It's the opposite: Your beloved Bayer sensors do crippled color
    sampling which is closer to bad nearest-neighbor effects than a
    full-color capture.

    > while a camera with a good AA
    > filter gives you an image much like the bicubic-resampled one. When
    > choosing a camera, I want the latter behaviour because it's predictable
    > and faithful to the original scene.


    You want the poor Bayer images with crippled color-sampling.

    Regards
    Guido
    Guido Vollbeding, Apr 2, 2004
    #8
  9. Guido Vollbeding <> writes:
    >Dave Martindale wrote:


    >> It is image content dependent. For some subjects, doing point sampling
    >> without any anti-alias filtering gives images that look *better* than if
    >> you'd done the theoretically correct thing.


    >The Foveon sensor does more area-sampling than point-sampling, and that's
    >not nearly so bad as you suppose.


    The SD9's fill factor is around 35%, so it is closer to point sampling
    than area sampling. The SD10's microlenses should give it a much better
    effective fill factor. But even a 100% fill factor, perfect area
    sampling, is just a box filter when it comes to reducing aliasing. A
    box filter just isn't very effective.

    >Wrong. SD9 and more SD10 do *area*-sampling, not point-sampling.
    >Therefore the comparison with nearest-neighbor resampling is wrong.


    Not far wrong. The SD9 samples 1/3 of the actual area of each pixel.
    The SD10 is probably closer to 100%, but this is just equivalent to a
    box filter in image processing. With a box filter, if you were reducing
    the size of an image by a factor of 2, you would average 4 input pixels
    together to produce one output pixel. In contrast, Photoshop's bicubic
    resampling filter should calculate each output pixel based on an 8x8
    array (64 total) of input pixels, and that's much more effective at
    removing aliasing.

    So the SD9/10 *are* somewhat better than point sampling, but still not
    as good as a camera that uses a proper anti-aliasing filter. And the
    difference between the Sigma cameras and most Bayer cameras (which do
    have such a filter) *is* similar to the difference between
    nearest-neighbour and bicubic downsizing in Photoshop. No, it's not
    exactly the same, but this was just an example of the difference.

    >It's the opposite: Your beloved Bayer sensors do crippled color
    >sampling which is closer to bad nearest-neighbor effects than a
    >full-color capture.


    This doesn't mean anything.

    >You want the poor Bayer images with crippled color-sampling.


    I want the best images I can get. If I *could* get 3-colour RGB images
    at full resolution and good uniformity with low noise, of course I'd
    prefer them. I *do* get such images from my flatbed scanner, and from
    a film scanner. Bayer images are inferior, but in well-understood ways
    that are mostly not visible in normal size prints. They are a
    compromise that works, most of the time.

    And Sigma images, despite having 3 samples per photosite, are even
    worse than Bayer for most subjects, due to noise, colour non-uniformity,
    and aliasing artifacts. If Bayer colour reproduction is "crippled", as
    you term it, then Foveon colour reproduction is missing several limbs
    completely. It's worse than Bayer, at least so far.

    Dave
    Dave Martindale, Apr 2, 2004
    #9
  10. David Kilpatrick

    John Navas Guest

    [POSTED TO rec.photo.digital - REPLY ON USENET PLEASE]

    In <c4j9om$431$> on Fri, 2 Apr 2004 08:54:46 +0000 (UTC),
    (Dave Martindale) wrote:

    >The SD9's fill factor is around 35%, so it is closer to point sampling
    >than area sampling. The SD10's microlenses should give it a much better
    >effective fill factor. But even a 100% fill factor, ...


    While a microlens array can greatly improve a low fill factor, it can't get
    anywhere near 100% in practice (due to how such lenses are made) -- at best no
    more than about 70-75%, or about double the 35%. So while the point sampling
    effect is reduced, it's not eliminated.

    --
    Best regards,
    John Navas
    [PLEASE NOTE: Ads belong *only* in rec.photo.marketplace.digital, as per
    <http://bobatkins.photo.net/info/charter.htm> <http://rpdfaq.50megs.com/>]
    John Navas, Apr 2, 2004
    #10
  11. David Kilpatrick

    Guest

    In message <>,
    Guido Vollbeding <> wrote:

    >Wrong. SD9 and more SD10 do *area*-sampling,


    Nonsense. The SD9 only samples 30% of the area within the rectangular
    active sensor area. Only soft optics will allow it to sample all the
    area of a scene.
    --

    <>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
    John P Sheehy <>
    ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
    , Apr 3, 2004
    #11
  12. David Kilpatrick

    Guest

    In message <>,
    Guido Vollbeding <> wrote:

    >You want the poor Bayer images with crippled color-sampling.


    They look much more realistic than the Foveon color, and that's the
    bottom line.
    --

    <>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
    John P Sheehy <>
    ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
    , Apr 3, 2004
    #12
    1. Advertising

Want to reply to this thread or ask your own question?

It takes just 2 minutes to sign up (and it's free!). Just click the sign up button to choose a username and then you can ask your own questions on the forum.
Similar Threads
  1. Dave Martindale

    Re: D30, 300D or SD9?

    Dave Martindale, Apr 1, 2004, in forum: Digital Photography
    Replies:
    22
    Views:
    724
    Ray Fischer
    Apr 11, 2004
  2. Chris Brown

    Re: D30, 300D or SD9?

    Chris Brown, Apr 1, 2004, in forum: Digital Photography
    Replies:
    11
    Views:
    521
  3. Bart van der Wolf

    Re: D30, 300D or SD9?

    Bart van der Wolf, Apr 1, 2004, in forum: Digital Photography
    Replies:
    10
    Views:
    536
  4. Mike Engles

    Re: D30, 300D or SD9?

    Mike Engles, Apr 1, 2004, in forum: Digital Photography
    Replies:
    4
    Views:
    301
    Chris Brown
    Apr 2, 2004
  5. Jordan Viray

    Re: D30, 300D or SD9?

    Jordan Viray, Apr 8, 2004, in forum: Digital Photography
    Replies:
    0
    Views:
    445
    Jordan Viray
    Apr 8, 2004
Loading...

Share This Page