[FFmpeg-devel] Another possible block-matching metric?

Jason Garrett-Glaser darkshikari
Tue Aug 21 07:05:23 CEST 2007


Correction, that should be (1-SSIM)*16000, since block-matching
metrics yield lower numbers for higher quality, not vice versa.

On 8/20/07, Jason Garrett-Glaser <darkshikari at gmail.com> wrote:
> This should be relatively easy to implement.
>
> I have been toying with x264 lately and found that SSIM can do a
> pretty good job as a block-matching metric if scaled appropriately.
> (1/(1-SSIM))*16000 seems to do pretty well, and gives similar results
> to SATD (Hadamard) in terms of PSNR and SSIM in the resulting encode
> and in some cases outperforms SATD by a small bit.  Adding SSIM as a
> block-matching metric to FFMPEG would be a nice encoding option;
> perhaps it will be just as effective for MPEG-2 and MPEG-4 ASP as it
> is for MPEG-4 AVC.  And since there's already assembly for it, it
> shouldn't need much coding at all.
>
> --Dark Shikari
>




More information about the ffmpeg-devel mailing list