Dismiss Notice
Attention AVForums app / Tapatalk users
Sadly GDPR means that, from 25th, we can no longer offer access to AVForums via the branded app or Tapatalk.
Click here for more information.

What's the point of FFDShow resizing?

Discussion in 'Desktop & Laptop Computers Forum' started by mikearv, Dec 5, 2004.

  1. mikearv

    mikearv
    Guest

    Products Owned:
    0
    Products Wanted:
    0
    Ratings:
    +0
    I've heard alot of people who say they set FFDShow to resize in Lanzos to resolutions higher than their display can handle to acheieve higher PQ.

    What exactly is the point of this when your DVD software & video card just resizes that back down again to at most, your max display resolution.

    Surely this is like taking a jpeg image that started life at 640x480, resizing it to 1024x768 then resizing it again to 800x600? If you have a graphics program and you compare a jpeg that's gone from 640x480->1024x768->800x600 to the same image that's gone 640x480->800x600 you will see the results are totally identical, so surely the same applies to video?

    Why use up all of that CPU power to perform FFDshow resize?
     
  2. KraGorn

    KraGorn
    Active Member

    Joined:
    Aug 30, 2003
    Messages:
    4,745
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    68
    Location:
    Warrington
    Ratings:
    +27
    The basic idea is that FFDSHOW is better at scaling than the graphics card and its' driver. Bearing in mind that up-sizing means filling in the gaps then there are a large number of ways of doing that, some produce better results than others, and FFDSHOW's Lanczos implementation is better than most.

    Now, I have to admit I'm not sure of the theory that says resize larger, except that it's probably the case that it's easier to resize to, say, twice the size since computers like multiples of 2 :) .. the algorithms I guess may work better than having to resize to a non-multiple and introduce fewer artifacts though I don't have the equipment capable of showing that.

    Down-sizing is a lot easier than up-sizing and tends to introduce far fewer artifacts so letting the graphics card down-size doesn't really impact PQ.


    At least, that's how I understand it. :)
     
  3. Mr.D

    Mr.D
    Well-known Member

    Joined:
    Jul 14, 2000
    Messages:
    11,167
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    133
    Ratings:
    +1,209

    Problem I have with the downsize technique is that I'm pretty sure most graphics cards use a bicubic for both upsize (which is fine) and downsize which is not : ideally you want to use a sinc type filter for downsizing. I've yet to figure out if graphics cards bother to employ different filtering for each process.

    Also you don't always employ the sharpest filter you can as often a sharper interpolation leaves you with more aliasing and beating than a softer "lesser" filter.
     
  4. mikearv

    mikearv
    Guest

    Products Owned:
    0
    Products Wanted:
    0
    Ratings:
    +0
    Sure but in the end the graphics card is scaling, something you cannot avoid?
     
  5. KraGorn

    KraGorn
    Active Member

    Joined:
    Aug 30, 2003
    Messages:
    4,745
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    68
    Location:
    Warrington
    Ratings:
    +27
    Yes, but the issue is what that process involves in terms of altering the image.

    Downsizing at its' simplest is simple pixel elimination whereas up-sizing means 'creating' pixels from adjacent ones, by definition this is adding information not present and thus artificial. It's how seamlessly the added pixels are integrated into the image that determines the resulting picture quality.

    Say you're up-sizing by a factor of 2, the easiest way is simply to do pixel doubling, which will look dreadful .. down-sizing by a factor of 2 at its' simplest means deleting every other pixel, which will be nowhere nearly as ugly.


    Mr. D:

    you may well be right, however many people do seem to re-size to 1440x960 so there must be something that makes it useful.
     
  6. Mr.D

    Mr.D
    Well-known Member

    Joined:
    Jul 14, 2000
    Messages:
    11,167
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    133
    Ratings:
    +1,209

    I'd dispute that to be honest. Some packages concatenate their image transform operators so a res up followed by a res down will counteract each other perfectly mathematically unless you actually render out each stage. (Shake is one package that does this ) Quite a few don't bother to retain the the information lost at each step so the differences from package to package are quite marked.

    I don't think the processes employed in getting a video res image to a higher display res will concatenate through ffdshow (certainly looking at processor overhead they don't).

    And of those packages that don't concatenate often whilst they use a decent default filter for upsizing they don't bother to change the filter for downsizing. (This is what actually worries me about the whole oversampling theory when applied to domestic PCs)

    And I could also argue that a video image should be completely linearised prior to any process that involves pixel value interpolation then reconverted back to video colourspace prior to display ( or left in linear and a suitable display LUT used to view it)
     
  7. Steve Bate

    Steve Bate
    Active Member

    Joined:
    Nov 10, 2002
    Messages:
    680
    Products Owned:
    0
    Products Wanted:
    0
    Trophy Points:
    21
    Location:
    Tarporley
    Ratings:
    +11
    Just when you thought you were beginning to understand :clown:
     

Share This Page

Loading...
  1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.
    Dismiss Notice