VideoHelp Forum




+ Reply to Thread
Results 1 to 13 of 13
  1. Member
    Join Date
    Jul 2002
    Location
    Israel
    Search Comp PM
    Clips captured at 704x576 I want to author as 352x576. Can I just encode the 704x576 avi into 352x576 mpg, or is there something more I should do?
    Quote Quote  
  2. The Old One SatStorm's Avatar
    Join Date
    Aug 2000
    Location
    Hellas (Greece), E.U.
    Search Comp PM
    Since it is 704 x 576, just load your source to your encoder and set the output direct to 352 x 576.
    This framesize is the easiest to downscale with minimum picture loss, since any encoder with bilinear resizing gonna drop each second horizontal "line" (dot, pixel, call it whatever you wish).
    La Linea by Osvaldo Cavandoli
    Quote Quote  
  3. Member mats.hogberg's Avatar
    Join Date
    Jul 2002
    Location
    Sweden (PAL)
    Search Comp PM
    Originally Posted by SatStorm
    drop each second horizontal "line"
    ...ahem - every second veritical column, you mean... ?

    /Mats
    Quote Quote  
  4. Originally Posted by SatStorm
    Since it is 704 x 576, just load your source to your encoder and set the output direct to 352 x 576.
    This framesize is the easiest to downscale with minimum picture loss, since any encoder with bilinear resizing gonna drop each second horizontal "line" (dot, pixel, call it whatever you wish).
    That's "nearest neighbor" resizing, not bilinear. And nearest neighbor is the worst possible resizing technique.
    Quote Quote  
  5. The Old One SatStorm's Avatar
    Join Date
    Aug 2000
    Location
    Hellas (Greece), E.U.
    Search Comp PM
    Convert 704 to 352 horizontal lines.
    The vertical lines remain 576

    That way you drop each even horizontal line (or call it whatever you wish)

    704 to 352 is the only scenario that nearest neighbor actually don't mess the picture.

    352 is exactly half 704
    La Linea by Osvaldo Cavandoli
    Quote Quote  
  6. Why drop every odd line when you can combine it with every even line? Seems to me a bilinear would be better than a nearest neighbor.


    Darryl
    Quote Quote  
  7. source: black white black white black white black white...

    half size nearest neighbor: black black black black black...

    half size bilinear: gray gray gray gray gray...
    Quote Quote  
  8. Originally Posted by junkmalle
    source: black white black white black white black white...

    half size nearest neighbor: black black black black black...

    half size bilinear: gray gray gray gray gray...
    Not sure what you are trying to say, but this will help illustrate why bilinear would be better. As you can see, either way you are losing information. But with bi-linear, you get a more accurate sampling of what should be there.


    Darryl
    Quote Quote  
  9. Member
    Join Date
    Aug 2002
    Location
    Sweden
    Search PM
    What about using lanczos resize or precice bicubic in this situation? Would that be worse than bilinear?

    I thnik it would be sharper at least but on the other side if you are encoding at low bitrates the quality may become better when using a soft resizing method.

    There are other things you can do also, like replacing borders that are hidden in the overscan area of the TV with black, and to do this optimal you should do it in sizes of 8 pixels (like 8,16,24 or 32 pixels borders).

    As an example:
    First crop your 704x576 source to 672x544.
    Then resize 672x544 to 336x544.
    Last you add 8 pixels of black borders to left and right side and 16 pixels black border on top and bottom.

    Now you have a 352x576 picture with black borders around thet you probably not will be able to see when watching it on TV.

    OK, but why doing this if I can't see it, you may say? Well why waste bits on things you can't see. You will get a little bit better quality at the same bitrate, or the same quality at a lower bitrate.
    Ronny
    Quote Quote  
  10. Originally Posted by dphirschler
    Originally Posted by junkmalle
    source: black white black white black white black white...

    half size nearest neighbor: black black black black black...

    half size bilinear: gray gray gray gray gray...
    Not sure what you are trying to say, but this will help illustrate why bilinear would be better. As you can see, either way you are losing information. But with bi-linear, you get a more accurate sampling of what should be there.


    Darryl
    Yes, that is what I was pointing out!
    Quote Quote  
  11. The Old One SatStorm's Avatar
    Join Date
    Aug 2000
    Location
    Hellas (Greece), E.U.
    Search Comp PM
    Whatever...

    I always use lanczos when I resize from more to less. And I always frameserve!

    BJ_M few years ago told me that tip while you use TMPGenc: Capture 704 x 576, encode at 352 x 576. TMPGenc resize only bilinear and 704 to 352 is the only scenario that TMPGenc actually does a good job on resizing horizontally.

    junkmalle said about nearest neighbor. I first wrote about bilinear. I don't even understand How I fucked up that post on later replies. Sorry (but those things happens when you work and answer the same time)
    La Linea by Osvaldo Cavandoli
    Quote Quote  
  12. Member
    Join Date
    Jul 2002
    Location
    Israel
    Search Comp PM
    Satstorm
    As in most cases, your replies seems to be most straight forward. Plus, from experience, your advise always work. Thanks.
    Quote Quote  



Similar Threads

Visit our sponsor! Try DVDFab and backup Blu-rays!