[Theora-dev] Re: [Fwd: Re: Theora mcomp tuning...]

[Adam D. Moss]

Timothy B. Terriberry wrote:
> Although SSD may seem to be a nicer theoretical quantity to work with, 
> this is only true when errors come from a Gaussian source.

Okay -- I guess that's true of most of my experiments on estimation
of perceptible quantization error.

 > Experiments
> have long shown that this is decidedly NOT the case. SAD in general 
> produces more reliable results, though finding a local error criterion 
> that is NOT subject to the occaisional gross estimation errors is 
> impossible, thanks to things like the aperture problem and repeated 
> structure.

One of the reasons I felt happier with SSD was that as far as I
could tell from a cursory read of the whole-frame error weight-
scaling (scan.c ?), SSD is also used in the qualititave assessment
of error there, so it seemed to make sense that these two methods
of assessment should be consistant.  But in the end, it seems like
quite a negligable difference either way.

Regards,
--Adam