Professional Web Applications Themes

Upscaling - Photography

"Hans-Georg Michna" <hans-georgNoEmailPleasemichna.com> wrote in message news:u0hvgvcg5h59lfg745moc8u5epmsjmmibt4ax.com... > 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 ...

  1. #1

    Default Re: Upscaling


    "Hans-Georg Michna" <hans-georgNoEmailPleasemichna.com> wrote in message
    news:u0hvgvcg5h59lfg745moc8u5epmsjmmibt4ax.com...
    > 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 Guest

  2. #2

    Default Re: Upscaling


    "Hans-Georg Michna" <hans-georgNoEmailPleasemichna.com> wrote in message
    news:t4n0hvcpr5re8dp2ubfq12s15gtbq6aki74ax.com...
    > "Bart van der Wolf" <bvdwolfnospam.nl> 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 Guest

  3. #3

    Default Re: Upscaling

    "Bart van der Wolf" <bvdwolfnospam.nl> wrote:
    >"Hans-Georg Michna" <hans-georgNoEmailPleasemichna.com> wrote in message
    >news:t4n0hvcpr5re8dp2ubfq12s15gtbq6aki74ax.com.. .
    >> "Bart van der Wolf" <bvdwolfnospam.nl> 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 Guest

  4. #4

    Default Re: Upscaling

    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" <bvdwolfnospam.nl> wrote:
    >
    > >"Hans-Georg Michna" <hans-georgNoEmailPleasemichna.com> wrote in message
    > >news:u0hvgvcg5h59lfg745moc8u5epmsjmmibt4ax.com.. .
    >
    > >> 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
    [email]staufferusfamily.net[/email]
    webpage- [url]http://www.usfamily.net/web/stauffer[/url]
    Don Stauffer Guest

  5. #5

    Default Re: Upscaling


    "Hans-Georg Michna" <hans-georgNoEmailPleasemichna.com> wrote in message
    news:osl2hvgjjgj39rpu6rt1rgrciaiifaipti4ax.com...
    SNIP
    > Just looking at their web site. Are you sure Qimage does vector
    > interpolation?
    Yes, I use it (almost) all the time:
    [url]http://www.ddisoftware.com/qimage/qimagehlp/interp.htm[/url]

    This is a page from *before* Vector interpolation was introduced:
    [url]http://www.ddisoftware.com/qimage/quality/[/url]

    This is an example (on one type of image):
    [url]http://www.ddisoftware.com/testpics/l-vs-v.jpg[/url]
    and of another type:
    [url]http://www.ddisoftware.com/qimage/vector.jpg[/url]

    Bart


    Bart van der Wolf Guest

  6. #6

    Default Re: Upscaling

    "Bart van der Wolf" <bvdwolfnospam.nl> wrote:
    >This is an example (on one type of image):
    >[url]http://www.ddisoftware.com/testpics/l-vs-v.jpg[/url]
    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 Guest

  7. #7

    Default Re: Upscaling

    Don Stauffer <staufferusfamily.net> 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 Guest

  8. #8

    Default Re: Upscaling


    "Hans-Georg Michna" <hans-georgNoEmailPleasemichna.com> wrote in message
    news:cjq4hvgdhgpgq5nf846u24mjtcrg0l46uj4ax.com...
    > "Bart van der Wolf" <bvdwolfnospam.nl> wrote:
    >
    > >This is an example (on one type of image):
    > >[url]http://www.ddisoftware.com/testpics/l-vs-v.jpg[/url]
    >
    > 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 Guest

  9. #9

    Default Re: Upscaling

    "Bart van der Wolf" <bvdwolfnospam.nl> wrote:
    >"Hans-Georg Michna" <hans-georgNoEmailPleasemichna.com> wrote in message
    >news:cjq4hvgdhgpgq5nf846u24mjtcrg0l46uj4ax.com.. .
    >> 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 Guest

  10. #10

    Default Re: Upscaling

    "Bart van der Wolf" <bvdwolfnospam.nl> 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 Guest

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139