Fun with various RD metrics

[Dark Shikari]

From "300" (all encoded in two-pass at 10 mbps, FGO was done using --fgo 10):

1. Magic new metric: framesize 92860 bytes:
http://thumbnails7.imagebam.com/701/cc65497000372.gif (http://www.imagebam.com/image/cc65497000372)

2. FGO metric (NSSD): framesize 106145 bytes:
http://thumbnails7.imagebam.com/701/f5e6ad7000375.gif (http://www.imagebam.com/image/f5e6ad7000375)

3. Normal metric (PSNR): 111552 bytes:
http://thumbnails7.imagebam.com/701/255b807000377.gif (http://www.imagebam.com/image/255b807000377)

From "Pirates of the Carribean" (all encoded in two-pass at 1 mbps, FGO was done using --fgo 10):

1. Magic new metric: framesize 1217 bytes:
http://img252.imageshack.us/img252/8658/new1217wy6.png

2. FGO metric (NSSD): framesize 1643 bytes:
http://img252.imageshack.us/img252/4962/fgo1643id6.png

3. Normal metric (PSNR): 1444 bytes:
http://img140.imageshack.us/img140/3117/old1444gh5.png

1 kilogram of SSD: 125 clocks
10 kilograms of SATD: 450 clocks
5 kilograms of SAD: 50 clocks
Using RD for detail retention: Priceless

[setarip_old]

@Dark Shikari

Hi!

To my decrepit old eyes, your "Magic new metric" images (especially the POC one) exhibits the loss of details in the shadows...

[Razorholt]

looks like you're adding artificial noise :)

[Dark Shikari]

looks like you're adding artificial noise :)Here's the original of the latter: it isn't artificial:

http://img133.imageshack.us/img133/8625/origqq5.png

This does make me think a subband method might be best though; try to match the amount of frequencies in each subband with the original source.

[Dark Shikari]

Its time for Test Metric 2! This time, I'll use a source where the difference is much more apparent.

Comments on parts of the image and the overall visual effect are appreciated; this is just another test metric among many. Clicking on the image will download the video.

Original:
http://i25.tinypic.com/vhtkes.png

Encoded (regular metric):
http://i28.tinypic.com/2i261k3.png (http://mirror05.x264.nl/Dark/force.php?file=./OriginalMetric.mkv)

Encoded (Test Metric 2):
http://i25.tinypic.com/33cu3ww.png (http://mirror05.x264.nl/Dark/force.php?file=./NewMetric.mkv)

As you can probably tell, the purpose of this RD metric is to reduce blurriness and sharpen the output, at the cost of a bit more potential ringiness.