Windows XP Incompatibilities - post here

[Dr Mordrid]

Brian,

Yup, the eye is very tolerant of chroma signal losses. This is why DV looks good to the eye. In fact it looks better than analog because of its proportional overabundance of luma, which the eye is much more sensitive to.

Problem is that MPEG encoders are not so tolerant.

Dr. Mordrid

[Flying dutchman]

Hi Doc,

with "Shifted" I meant that the 8x8 macroblocks of a digital camera no longer coincide with the MJPG macroblocks because the capture resolution is simply different. Like in a bilinear re-scaling of 720x576 to 704x576.


I don't have a DV card (yet) but I'd be grateful if you would try out the following experiment:

- capture DV over firewire
- In virtualdub, shift the picture 4 pixels to the right and 4 pixels downward
- then crop the image to 704x576 and encode to MJPG.

This manipulation ensures that the DV macroblocks no longer coincide with the MJPG macroblocks; the "edges" of the DV macroblocks are now exactly in the centre of the MJPG macroblocks. I am eager to find out if this re-aligning is enough to give the same "improvement" as analog capturing; I'd like to keep as much bandwidth as possible...

[Dr Mordrid]

Remember I'm working in NTSC and so use a 480 high frame.

When compositing 704 video in a 720 wide frame I just apply a custom 2D moving path created for the purpose.

This custom path horizontally centers 704 wide frames to the the 720 wide frame, which leaves an 8 pixel (1 macroblock) wide border on each side. These borders are invisible when the finished product is played on a TV as they are in the overscan region.

Since DV and MJPeg both use 8x8 macroblocks they are aligned as long as you either place the MJPeg clip at either horizontal extreme or center it to the 720 wide frame.

I see no reason to shift it vertically since both are 480 high.

HuffYUV gets the same treatment even though it's not DCT based and therefore has no macroblocks. For me an 8 pixel border on each side is just more esthetically pleasing than a 16 pixel border on one side or the other.

Dr. Mordrid

[Flying dutchman]

Hi Doc,

my idea was to deliberately de-align the macroblocks by shifting the DV image 4 pixels both horizontally and vertically. The shifting process moves the crosspoints of 4 adjacent macroblocks to the exact centre of the new macroblocks.
I'd simply like to find out if this misaligned image re-compresses nicer than the original DV image, because I hear that lots of people get disappointing results when converting DV to MPG or MJPG. I am going to process some DV tapes shortly, and I'd rather not capture them the analog way because:

1) DV already has the 16-235 luminance range needed for (VCD/SVCD/DVD) MPG, so I might as well avoid the cumbersome calibration needed for the Marvel.

2) I'd like to keep as much bandwidth inside and noise outside as possible.

I realize that DV only has 4-1-1 colorspace but re-capturing it the analog way doesn't change that, it is just 4-1-1 re-compressed to 4-2-2, isn't it?