SSIM value interpretation

[liquidator87]

From what I've read, the SSIM value is the most suitable to quantify the "quality" of the encode... are there any values which discriminate, for example, a "very good" encode from a "good" one, or a "medium" encode from a crappy one? Or maybe I just have to read it as "the closer to 1, the better is"?

[juGGaKNot]

Your eyes are the most suitable to quantify the "quality" of the encode.

[Dark Shikari]

It's probably a reasonable interpretation that 0.98 is twice as good as 0.96 is twice as good as 0.92... etc. There's no magic number though, and I've seen sources go as low as ~0.88 SSIM without significant visual quality loss, while others looked bad at 0.98.

[liquidator87]

@Dark Shikari: thank you

[LoRd_MuldeR]

liquidator87, I recommend you also have a look at the original SSIM paper:
http://www.cns.nyu.edu/pub/eero/wang03-reprint.pdf

Another paper about SSIM (as well as other quality metrics) that I found quite interesting:
http://www.ece.uwaterloo.ca/~z70wang/publications/SPM09.pdf

[liquidator87]

Very interesting reading, especially the second one