[Open-graphics] ATI and Nvidia raise the bar?
James Richard Tyrer
tyrerj at acm.org
Sun Sep 17 19:46:15 EDT 2006
Dieter wrote:
> So ATI and Nvidia now have noise and sharpness filters.
> Perhaps I missed it, but I didn't see anything about whether this
> is done in hardware or in the driver. I also haven't seen an
> explaination of exactly what this "twinkling" is and what causes
> it.
>
> http://www.behardware.com/articles/635-1/ati-and-nvidia-corrects-the-twinkling-effect-of-lcds-in-movies.html
IIUC, the "twinkling" is caused by the MPEG artifacts and this is
accentuated when using some flat panel screens.
The PSEUDO-noise reduction and ARTIFICIAL sharpening are actually the
same function. A small two dimension FIR filter is applied to the image.
I emphasize that this is pseudo-noise reduction because true noise
reduction requires taking the FFT of the data, modifying the FFT by
attenuating the high spatial frequencies or using a procedure developed
by Bell Labs which requires multiple operations on the FFT, and then
taking the inverse FFT. This procedure is used to remove noise from
images and audio but doing it in real time at 30 or (worse) 60 fps would
require a super computer.
With artificial sharpening, only the effect of greater sharpness is
produced by increasing local contrast without changing global contrast.
IIUC, these two operations might be somewhat in opposition to each other
except that noise has higher spatial frequencies than the edge sharpening.
IAC, since this involves running the image data through two filters, you
can accomplish the same thing by running it through one filter which is
the normalized product of the two.
This is going to be quite computationally expensive since an 8x8 filter
is going to require 64 MAC operations for each pixel in the image. And,
you would really need a larger filter to avoid the 8x8 block artifacts.
But a 5x5 (5 and 8 are mutually prime) might be sufficient -- that is
still 25 MACs per pixel.
I have to wonder exactly how this is being done.
--
JRT
More information about the Open-graphics
mailing list