Whoa, huge SSIM boost from deen()

[Sagekilla]

Okay, so I finally decided to try out denoising after looking at the topped thread on it and my first reaction was "Whoa, how the hell did that happen?"

http://img.photobucket.com/albums/v621/Sagekilla/SSIM_boost.png

Top image uses deen, bottom without it. I'm using --crf 20.5, which might have something to do with that.. But I mean is this seriously normal? I know SSIM doesn't increase linearly, but this isn't a tiny increase at all: SSIM went up by 0.09 points! Even PSNR went up too.

Any explanations are greatly appreciated!

[Manao]

If you do the same, but replace deen() by blankclip(), you get a SSIM of 1.0 !

That's the problem : SSIM must be computed against the unfiltered source.

[Unearthly]

I believe what is happening is that Deen is removing noise (and possibly detail, depending on how strong your settings are) and this makes your frames easier to encode. Now remember that SSIM is a metric that compares your original frame to your encoded frame. So your frame is easier to encode, thus staying closer to the original, which explains the increase in SSIM.

This doesn't mean it's higher quality, it just means that less was lost while encoding. If you replaced your whole clip with a black screen I bet SSIM would go up as well. :p

[Dark Shikari]

You'll get even more using FFT3DFilter(). Removing noise makes it far, far easier to compress a frame, so x264 will give much much better results without that noise. It doesn't mean its closer to the original, but it does mean its higher quality relative to what you inputted to the encoder.