Re: Upscaling

Discussion in 'Digital Photography' started by Bart van der Wolf, Jul 12, 2003.

  1. "Hans-Georg Michna" <> wrote in message
    news:...
    > Many thanks for the good hints!
    >
    > Does anybody here understand how the upscaling algorithms like
    > Lanczos interpolation work? Any experiences?


    Do a Google search for a message with a description that Dave Martindale
    gave in this forum (much clearer than I could explain it) some months (?)
    ago.
    "Lanczos windowed sinc" interpolation is very good in retaining the original
    pixel contrast when looked at from an appropriate distance. It can cause
    some visible (ringing) artifacts on sharp edges if looked at from too close.
    Other interpolation methods (such as bi-cubic) can cause loss of pixel
    contrast which also looks less than stellar, so it depends on image content
    what method is better for a give output size. Grainy/noisy images may even
    benefit from some mushifying.
    Programs like Qimage can also use a different "Vector interpolation" method
    that works very good for many types of images.

    Bart
    Bart van der Wolf, Jul 12, 2003
    #1
    1. Advertising

  2. "Hans-Georg Michna" <> wrote in message
    news:...
    > "Bart van der Wolf" <> wrote:

    SNIP
    > >Programs like Qimage can also use a different "Vector interpolation"

    method
    > >that works very good for many types of images.

    >
    > Bart,
    >
    > I'm not actually looking for any kind of interpolation. What I
    > want is a program that detects high-contrast edges and redraws
    > them sharply in the higher resolution, which would amount to a
    > kind of anti-aliasing function.


    Try the Qimage Vector interpolation, it "invents" plausible edges, without
    the posterized look of Genuine Fractals (and at a better price).

    Bart
    Bart van der Wolf, Jul 13, 2003
    #2
    1. Advertising

  3. "Bart van der Wolf" <> wrote:

    >"Hans-Georg Michna" <> wrote in message
    >news:...


    >> "Bart van der Wolf" <> wrote:


    >> >Programs like Qimage can also use a different "Vector interpolation" method
    >> >that works very good for many types of images.


    >> I'm not actually looking for any kind of interpolation. What I
    >> want is a program that detects high-contrast edges and redraws
    >> them sharply in the higher resolution, which would amount to a
    >> kind of anti-aliasing function.


    >Try the Qimage Vector interpolation, it "invents" plausible edges, without
    >the posterized look of Genuine Fractals (and at a better price).


    Bart,

    thanks, will look. "Invents plausible edges" sounds exactly like
    what I'm hoping for.

    [Time passes.]

    Just looking at their web site. Are you sure Qimage does vector
    interpolation? The site only mentions Lanczos interpolation, not
    a word about anything else. They even have examples that show a
    very smooth interpolation, but no edge sharpness.

    Hans-Georg

    --
    No mail, please.
    Hans-Georg Michna, Jul 13, 2003
    #3
  4. Bart van der Wolf

    Don Stauffer Guest

    When you say redraws it in 'higher resolution', that is implying a
    resampling or interpolation. Sharpening per se does not necessarily
    lead to higher res. It CAN, if the actual resolution of original image
    is well below Nyquist limit, but not if it is at limit.

    Most sharpening filters are the equivalent of a differentiation added to
    a proportional function.

    Hans-Georg Michna wrote:
    >
    > "Bart van der Wolf" <> wrote:
    >
    > >"Hans-Georg Michna" <> wrote in message
    > >news:...

    >
    > >> Does anybody here understand how the upscaling algorithms like
    > >> Lanczos interpolation work? Any experiences?

    >
    > >Do a Google search for a message with a description that Dave Martindale
    > >gave in this forum (much clearer than I could explain it) some months (?)
    > >ago.
    > >"Lanczos windowed sinc" interpolation is very good in retaining the original
    > >pixel contrast when looked at from an appropriate distance. It can cause
    > >some visible (ringing) artifacts on sharp edges if looked at from too close.
    > >Other interpolation methods (such as bi-cubic) can cause loss of pixel
    > >contrast which also looks less than stellar, so it depends on image content
    > >what method is better for a give output size. Grainy/noisy images may even
    > >benefit from some mushifying.
    > >Programs like Qimage can also use a different "Vector interpolation" method
    > >that works very good for many types of images.

    >
    > Bart,
    >
    > I'm not actually looking for any kind of interpolation. What I
    > want is a program that detects high-contrast edges and redraws
    > them sharply in the higher resolution, which would amount to a
    > kind of anti-aliasing function.
    >
    > Hans-Georg
    >
    > --
    > No mail, please.


    --
    Don Stauffer in Minnesota

    webpage- http://www.usfamily.net/web/stauffer
    Don Stauffer, Jul 13, 2003
    #4
  5. Bart van der Wolf, Jul 13, 2003
    #5
  6. "Bart van der Wolf" <> wrote:

    >This is an example (on one type of image):
    >http://www.ddisoftware.com/testpics/l-vs-v.jpg


    Bart,

    thanks, I had missed those at first.

    But this isn't what I expected. It is quite good, but I don't
    understand why the algorithm makes a perfectly sharp image
    unsharp and fuzzy. It shouldn't do this, I think.

    The desired result would be an arrow that looks exactly like the
    first one, except without the steps.

    Hans-Georg

    --
    No mail, please.
    Hans-Georg Michna, Jul 14, 2003
    #6
  7. Don Stauffer <> wrote:

    >When you say redraws it in 'higher resolution', that is implying a
    >resampling or interpolation. Sharpening per se does not necessarily
    >lead to higher res. It CAN, if the actual resolution of original image
    >is well below Nyquist limit, but not if it is at limit.
    >
    >Most sharpening filters are the equivalent of a differentiation added to
    >a proportional function.


    Don,

    I know, but I'm not actually looking for sharpening. I'm looking
    for anti-aliasing, taking the pixel steps out when scaling up.

    It seems though that the function I'm looking for does not exist
    yet as a readily available piece of software.

    Let me demonstrate the function with just black and white
    pixels.

    Picture before upscaling (4 symbols = 1 pixel):

    ##
    ##
    ####
    ####
    ######
    ######
    ########
    ########

    Picture after upscaling with factor 2 (1 symbol = 1 pixel):

    #
    ##
    ###
    ####
    #####
    ######
    #######
    ########

    It would need something like edge detection and edge redrawing.

    I think the video upscaling chips (Faroudja etc.) do exactly
    this. I can see it every day in my home cinema. When the picture
    moves rapidly, the chip is too slow, but as soon as the picture
    becomes less dynamic, the edges are smoothed.

    Hans-Georg

    --
    No mail, please.
    Hans-Georg Michna, Jul 14, 2003
    #7
  8. "Hans-Georg Michna" <> wrote in message
    news:...
    > "Bart van der Wolf" <> wrote:
    >
    > >This is an example (on one type of image):
    > >http://www.ddisoftware.com/testpics/l-vs-v.jpg

    >
    > Bart,
    >
    > thanks, I had missed those at first.


    No, it wasn't mentioned on the web site, but in the Yahoo support forum, so
    you couldn't know.

    > But this isn't what I expected. It is quite good, but I don't
    > understand why the algorithm makes a perfectly sharp image
    > unsharp and fuzzy. It shouldn't do this, I think.


    Because the program has to guess what is image detail, and what is a pixel
    boundary.

    > The desired result would be an arrow that looks exactly like the
    > first one, except without the steps.


    If the first one would have been rotated by just a few degrees, all edges
    would have pixel steps (aliasing), whereas Vector interpolation would have
    made a (better) compromise. Then some unsharp masking would produce a
    visually superior result.

    Bart
    Bart van der Wolf, Jul 14, 2003
    #8
  9. "Bart van der Wolf" <> wrote:

    >"Hans-Georg Michna" <> wrote in message
    >news:...


    >> But this isn't what I expected. It is quite good, but I don't
    >> understand why the algorithm makes a perfectly sharp image
    >> unsharp and fuzzy. It shouldn't do this, I think.


    >Because the program has to guess what is image detail, and what is a pixel
    >boundary.


    >> The desired result would be an arrow that looks exactly like the
    >> first one, except without the steps.


    >If the first one would have been rotated by just a few degrees, all edges
    >would have pixel steps (aliasing), whereas Vector interpolation would have
    >made a (better) compromise. Then some unsharp masking would produce a
    >visually superior result.


    Bart,

    I guess it's a matter of taste. The original picture didn't have
    any unsharp masking (or anti-aliasing using intermediate
    pixels). Why should the upscaled one have it?

    But this discussion may not make much sense now. Nobody will
    listen to me and write a new program. :)

    Hans-Georg

    --
    No mail, please.
    Hans-Georg Michna, Jul 16, 2003
    #9
  10. "Bart van der Wolf" <> wrote:

    >In this, and only situations like it (exact integer multiple upscaling of
    >sharp edges), a so called bi-linear interpolation would have done just that.
    >It demonstrates that the image content (and scale factor) determine which
    >interpolation method does best.


    Bart,

    are you sure? I would have thought it would produce some grey
    pixels.

    But anyway, I would like an algorithm that repaints edges,
    perhaps even if they were slightly unsharp in the original.

    Hans-Georg

    --
    No mail, please.
    Hans-Georg Michna, Jul 16, 2003
    #10
    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. Hans-Georg Michna

    Upscaling Programs - Qimage, Genuine Fractals, pxl SmartScale

    Hans-Georg Michna, Oct 25, 2003, in forum: Digital Photography
    Replies:
    16
    Views:
    11,886
    Hans-Georg Michna
    Oct 26, 2003
  2. Doug Jacobs

    Upscaling DVD player reccomendations

    Doug Jacobs, Jan 29, 2007, in forum: DVD Video
    Replies:
    4
    Views:
    650
    ThePunisher
    Jan 31, 2007
  3. Replies:
    2
    Views:
    526
Loading...

Share This Page