summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorSøren Sandmann Pedersen <ssp@redhat.com>2011-04-05 12:56:57 -0400
committerSøren Sandmann Pedersen <ssp@redhat.com>2013-07-29 06:21:50 -0400
commit7d0d0de6bf534d3b8df3b4f542242154dc2998a2 (patch)
treee198e64bf4f808ec1ac86b564308291f9a65115c
parentef89fff8da1429784209124468fc92af569d13fb (diff)
More dither
-rw-r--r--docs/yuv-dither.txt12
1 files changed, 11 insertions, 1 deletions
diff --git a/docs/yuv-dither.txt b/docs/yuv-dither.txt
index fa8cd22f..c176ce95 100644
--- a/docs/yuv-dither.txt
+++ b/docs/yuv-dither.txt
@@ -21,7 +21,17 @@ Once we have such a box, generate triangular noise and store it as
pixels.
Then, use void-and-cluster algorithms to make the Y/U/V noise jointly
-blue.
+blue. This can be done separately for each plane.
+
+A way to look at this operation might be, instead of using
+void/cluster, simply try to minimize the perceptual impact by
+filtering each channel with a perceptually motivated filter and then
+swap pixels until that filter is as close to zero as possible. This
+sidesteps the issue of gray-levels.
+
+The noise quality of doing this may not be good enough though, so it
+could still be useful to do the whole graylevel optimization.
+
Finally, convert to RGB. (Maybe do subpixel sampling?)