[MEncoder-users] capture: synchronization problems

[D Richard Felker III]

On Wed, Mar 23, 2005 at 10:59:34AM -0800, Phil Ehrens wrote:
> Kyle Schmitt wrote:
> > > I've never seen such artifacts. I could be wrong
> > > though. Could someone tell me 
> > > exactly what artifacts? MJPEG by definiton is lossy,
> > > but for all my capturing 
> > > needs I found it very high quaality at small
> > > filesize.
> > Ahh here's the rub.  Although both mjpeg and mpeg1/2/4
> > are lossy codecs, the artifacts they produce are
> > different.  <oversimplification>Mpeg codecs simplify
> > the image into blocks that can be shifted.  The fewer
> > blocks it uses, and the more blocks it reuses, the
> > smaller the file.</oversimplification>  <gross
> > oversimplification>Jpeg translates the image into
> > math, and then saves space by simplifying the math NOT
> > the image.  The more the math is simplified the
> > smaller the file.</gross oversimplification>
> 
> This implies there is some fundamental difference
> between mpeg and jpeg. Mpeg is VERY close to jpeg
> in terms of image encoding, but mpeg uses interframe
> coding techniques that encode only differential
> changes in the frames between the keyframes.
> 
> A good example of where this becomes highly efficient
> and compresses well would be a scene of a huge field
> of snow with a tiny figure moving across it. Only the
> macroblocks containing the figure change between frames.
> In some cases this results in artifacts that make the
> figure look like it is moving in a fuzzy bubble. There
> are all sorts of other artifacts that are possible.

This only happens if you use bad quantization, and will happen with
either mpeg or mjpeg unless you use sufficient bitrate or fixed low
quantizer.

Rich