Announcement

Collapse
No announcement yet.

Best Process for VCD's? (again)

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Best Process for VCD's? (again)

    Hi All!
    OK...am really learning tons and am at the point where I'd love to revisit this old issue. Am interested in making the absolute best vcd's possible so would welcome any tips/hints/etc from any of you.

    As a starting point, here is the process that am currently using:

    1) Capture- Avi-i/o using the following settings...352x480, 29.97fps, 44.1KHz stereo, uncompressed. Comment: The quality is great, the files are huge and for some reason am dropping a couple frames per minute. Some suggest this is due to the difference between 30 and 29.97fps but would love to eliminate them as wading through an hour of video frame by frame to eliminate the drops is painful to say the least.

    2) Editing/Output- Adobe Premiere- Import 2,096meg raw capture video segments. Go through and edit out all the drops, edit down a bit, add transitions, apply clip filter as necessary. Export movie using LSX plug-in using VCD settings.

    The final results are excellent but can't help but feel that better results are possible.

    Would love to eliminate those dropped frames on the raw capture and improve the clarity/sharpness of the final video as much as possible. Have read in this forum about Temporal Cleaning, Noise Filters and a slew of other things that I really am clueless about.

    Would love to hear from others as to their 'best practices'!

    Thanks as ever!!!!
    -funsoul

    My system:
    - pIII 500eb (not overclocked)
    - 256meg ram
    - asus p3bf mobo
    - Windows 98se
    - 40g maxtor primary
    - Promise Fastrack66 w/2 27.3g ibm deskstar gxp drives
    - matrox g400tv
    - TurtleBeach MontegoII+

    All captures are from videotapes.
    mmedia pc: 2x2.4/533 xeons@3.337ghz, asus pc-dl, 2g pc3500 ddram, 27g primary, 2x120 WD's, promise fastrack100, matrox g400-tv, hercules soundcard Server box: p4 1.4GHz, asus p4t, 1g ecc rdram, 27.3g primary, 3x80g maxtors, promise fastrack66, radeon ve, soundblaster Beat box: p3 500, asus p3bf6, 1024meg pc100, 45g primary, 3x45g maxtors, soundblaster, radeon ve, dazzle vcII

  • #2
    My preferred setup(PAL, not NTSC):

    Capture MJPG at full-resolution 704x576 using Avi_io. Full resolution images, properly de-interlaced and scaled down vastly improve signal/noise ratio.

    Load all file segments into VirtualDub.

    Apply the following Filters:
    If necessary, "temporal cleaner".
    "smart deinterlace" using default settings.
    Crop the black borders away.
    Resize (precise bicubic) to 352x288

    Start "Frameserver" mode
    Use tMpegencoder 1.24 beta 12 for encoding. Use the "PAL VCD" template, make sure motion estimation is NORMAL (because higher settings make the picture more pale-must be still a bug?)

    Resistance is futile - Microborg will assimilate you.

    Comment


    • #3
      Originally posted by Flying dutchman (but snipped viciously by myself):
      My preferred setup(PAL, not NTSC):
      Crop the black borders away.
      Resize (precise bicubic) to 352x288

      Use the "PAL VCD" template,
      I admit to being slightly lost here. If the borders are cropped away, and the image resized to the correct PAL VCD settings, doesn't that mean that parts of the image (that had been taken up by the black borders) will now be lost off the edge of the tv screen? Or am I misunderstanding how it work?

      And can PAL VCD have a 16:9 aspect ratio, or is it fixed at 4:3?

      [This message has been edited by Meeper (edited 18 September 2000).]

      Comment


      • #4
        Flying dutchman: What settings in TMPGenc do you use for Interlace?
        There is one dropdown box to choose Frame Type: Progressive/Interlaced,
        but then a check box that says "Deinterlace (none)" which is
        confusing - do you check it to _not_ deinterlace? (Mayve a bad
        jp->en translation). Do you do anything with Quantizer Matrix?
        I capture at 352x480 because I "only" have 9G of capture space,
        and one hour of highest quality mjpeg will fill that.
        What settings for temporal cleaner do you use? What I get are
        little "sparkles" due to bad TV reception that I'd love
        to be able to filter out.

        Comment


        • #5
          Frankie:
          the deinterlacing is in the filter section
          (use smart de-interlace).

          Flying Dutchman:

          1)
          temporal cleaner results for me in shadow effects after mpg-encoding, because of the following reason (I think): suppose an object is moving in front of a uniform background. If in a certain fram at a certain position you have the object, but in the folliwing frame there is background, there is no noise reduction at that position (because there was no background there in the earlier frame). This noice translates in shadow effects after mpg-encoding.
          Do you see the same artefacts ?

          2)
          I think I'm using the latest version of tMpeg, and I don't get it to work with the frameserver mode of virtual dub (this was my first experiment with the frameserver mode, though). I think I followed the instructions (shouldn't be too hard I guess).
          Any hint or trick ?

          3)
          I'm also experimenting a little bit with SVCD, using more or less your VCD approach, but resizing to 480x576 (PAL). Since the framserver mode does not work for me, I have to save again the resulting video file, but the MJPEG compressor gives an error message, only if I apply the resizing.
          Any ideas ?
          AMD-K6-2-300, EPOX-58MVP3C-M,
          Marvel G400, 64MB+64MB(PC100+PC133), Maxtor DM40PLUS 30GB,
          W98 (no SE), currently no overclocking (FSB=100)

          Comment


          • #6
            2) Frameserver: I had trouble with it at first but I think I ran
            the proxyon.reg file and since then it works OK. I'm using
            Windows 98. I save the second file with an .avi extension
            and TMPGenc as well as Ligos read it just fine. Although
            it takes me 10 hours to encode 22 minutes of MPEG-1 with TMPG....

            Deinterlacing: It can be done in either Vdub or TMPG,
            which is better? I currently feed 352x480 into TMPG and let
            him do it.

            Comment


            • #7
              Frankie,

              thanks for the tip. Where to find that proxyon.reg file (I'm not at my homPC currently) ?

              a) are you sure tMpeg really does de-interlacing, or does it just encode the odd & even frames separately, maintaining the interlaceing?

              b) are you pleased with the result. I have difficulties to believe that tMpeg would do as well in de-interlacing as does the plug-in for virtualdub.
              AMD-K6-2-300, EPOX-58MVP3C-M,
              Marvel G400, 64MB+64MB(PC100+PC133), Maxtor DM40PLUS 30GB,
              W98 (no SE), currently no overclocking (FSB=100)

              Comment


              • #8
                Frankie,
                in the TMPGEnc the Interlaced/Non-Interlaced key (the older translation says Progressive/Interlaced) refers to the Video source type.
                Doubbleclick on the Deinterlace filter and make your choice between a lot of deinterlacing methods.
                It works.


                Fred H
                It ain't over 'til the fat lady sings...
                ------------------------------------------------

                Comment


                • #9
                  Thanks Flying Dutchman!

                  Well...gee...between you and Eric now have a ton of new tricks n' techniques to try along with an almost infinite variety of variables...am sure it'll keep me busy for a while *grin*

                  Just wanted to say thanks!

                  Ummmm....one more question, though (seems like I always say that these days!). Always thought that if one cuts away from the top/bottom of the video (using the clip filter as opposed to the crop feature in premiere) that resizing it to fit would actually reduce the final quality (as the clipping reduces the size and the resize expands it). If that's the case, would prefer to keep the quality and just deal with the black bordering. Your thoughts?
                  mmedia pc: 2x2.4/533 xeons@3.337ghz, asus pc-dl, 2g pc3500 ddram, 27g primary, 2x120 WD's, promise fastrack100, matrox g400-tv, hercules soundcard Server box: p4 1.4GHz, asus p4t, 1g ecc rdram, 27.3g primary, 3x80g maxtors, promise fastrack66, radeon ve, soundblaster Beat box: p3 500, asus p3bf6, 1024meg pc100, 45g primary, 3x45g maxtors, soundblaster, radeon ve, dazzle vcII

                  Comment


                  • #10
                    I absolutely hate borders of anything around videos I make (black or otherwise). However, I don't resize upwards, only downwards, so I only have guesses to offer on that front (when I resize down to 448x336, it looks immensely similar whether I had to crop 2 pixels from the bottom, or 20). I recommend taking a few frames with lots of detail and checking how they look when cropped and re-scaled as opposed to left alone (make sure you use a useful resizing scheme, such as bicubic). My guess is that you will not find it noticable until the cropped segments are more than about 5% of the image. (If you use a skanky resizing scheme like "nearest neighbor" then I guarantee you will notice some problems with it).

                    Comment


                    • #11
                      [QUOTE]Originally posted by MarkD:
                      1)
                      >temporal cleaner results for me in shadow >effects after mpg-encoding, because of the >following reason (I think): suppose an >object is moving in front of a uniform >background. If in a certain fram at a >certain position you have the object, but >in the folliwing frame there is background, >there is no noise reduction at that >position (because there was no background >there in the earlier frame). This noice >translates in shadow effects after mpg->encoding.
                      >Do you see the same artefacts ?

                      This sounds odd... Most likely what is happening is that there is some pixellocking or blending from the moving object and the background. My best advice is to try reducing the thresholds. (also make sure you are using the latest version.. I fixed a lot of shadow issues in the latest build)

                      Comment


                      • #12
                        Yeah, resizing does cost some quality (a
                        tiny little bit). "precise bicubic" does a pretty good job, though. If you prefer not to, at least try to make the black border centered in equal thickness around the whole image.

                        I've made an interesting observation though: The width of the vertical black border at the left-hand side is depending on the video signal. If I feed a VHS tape into my Marvel, it's a lot thicker than if I connect my DV camcorder. So, theoretically, if I would delay the Hsync impulse of the video signal just a tiny little bit, it should be gone... Anybody out there own a video processor and care to try?

                        About "temporal cleaner": It's my experience that it makes a difference where you put it in the filter chain (which is better, before or after de-interlacing?)


                        One last trick: M$ mpeg-4 codec does an impressive job at filtering. I have made several good VCD's where I used this format as an intermediate storage format.(just applied smart de-interlace). Leave all filtering disabled in your mpeg encoder!


                        Resistance is futile - Microborg will assimilate you.

                        Comment


                        • #13
                          Flying Dutchman--
                          I wold have to agree with you on the MPEG4, I hadn't used it fro awhile, but I started palying with it again last night. The quality is great. I captured Huff, then used VDub to recompress it. It is so much faster, smaller, cleaner and you can scroll it in Media Player with no problem. I just wish that I could use it to make (s)vcd's for a standalone. Oh well, maybe I'll build something to throw in my A/V rack.
                          WinXP Pro SP2 ABIT IC7 Intel P4 3.0E 1024M Corsair PC3200 DCDDR ATI AIW x800XT 2 Samsung SV1204H 120G HDs AudioTrak Prodigy 7.1 3Com NIC Cendyne DVR-105 DVD burner LG DVD/CD-RW burner Fortron FSP-300-60ATV PSU Cooled by Zalman Altec Lansing MX-5021

                          Comment


                          • #14
                            -----------------------------------
                            About "temporal cleaner": It's my experience that it makes a difference where you put it in the filter chain (which is better, before or after de-interlacing?)
                            -----------------------------------

                            I used to put it after the de-interlacing ("gotta deinterlace first because that's the worst problem" mentality), but I've lately taken to putting the cleaner as the very first filter, as it seems to work slightly (very slightly) better at that point. I've experimented with two TC filters, one before and one after, with different settings, and those alternately give me better or worse results depending on the video. There's no quality AI better than a human watching the result

                            Comment


                            • #15
                              As for where the best place for temporal cleaner... I usually put it as the very last filter (after a resize). I usually find that after you do an operation that reduces spatial noise (most resize ops will reduce some spatial noise), it's time to reduce temporal noise, which usually knocks the most noise out of the signal. This is all IMHO of course. One other VERY important thing to do is to experiment with the filter settings to find the best settings for the job. Use the show motion option to see exactly what any set of settings is doing to the image.
                              Anyways, when/if I get the new temporal cleaner is finally stable, it should be a little bit better at noise reduction.

                              Comment

                              Working...
                              X