Software Scaling sharpness

[Ryan Koo]

I searched this forum but couldn't find a thread about this specifically. Shoot me if it's been covered a hundred times, but...

I noticed that a 1080p output from Resolve was sharper compared to the same shot exported to 1080p from REDCINE-X. In the latter under Export settings the Software Scaling options are numerous -- there are ten of them and it defaults to Mitchell. After choosing some of the sharper settings there was a marked improvement. Anyone have any experience with these? I tested several of them on one shot and found myself liking Lanczos3, but that doesn't mean much to me and they're not covered in the REDCINE manual at all. There's a screenshot of the options attached, and I also made an animated gif (blown up to 300%) comparing Mitchell (softer, the default) vs. Lanczos3 (sharper). I know a lot of us are doing 1080p outputs for obvious reasons so thought this was interesting...

Actually, the .gif doesn't animate on REDUSER, so use this link:

http://nofilmschool.com/files/images...re_scaling.gif

Look at the ADIDAS lettering... thoughts?

[Shane Kelly]

good question, I'd like to know the answer as well.

[Ivan Kovax]

If you outputted via red rocket, there is no added sharpness there...

[ericyoung]

The scaling algorithms prioritise different things. Generally sharper algorithms, while obvously looking "sharper", will tend to have more visible artifacts like halos, aliasing or ringing on certain picture elements such as high contrast edges or colour boundaries (eg text) or high frequency elements. Also, some will exaggerate noise, so would not be as useful on noisy footage. Some will be better overall quality, but take longer to process if not hardware accelerated. Some are better when scaling by a large ratio. Or are better at upscaling than downscaling. And so on. Having a choice of algorithms lets us choose the best fit for our footage, computing power, deadlines and intended deliverables.

Lanczos is a good general purpose filter which balances sharpness with processing time, and minimal artifacts for the majority of well exposed footage, but unfortunately there is no real alternative to trial and error with your particular footage. Just test short representative sections of your footage before doing a massive render!

Googling "video scaling algorithm" doesn't seem to bring up an overall comparison of all the available algorithms unfortunately, although this book might be a good reference:
http://my.safaribooksonline.com/book...ing_algorithms