Probst article in AC - questions about Nyquist Limit and Red resolution

[Mark Wilkinson]

I just read Christopher Probst's article in AC this month "Decoding Digital Imagers: Part 1" which is an extremely cogent explanation of digital sensors and what's going on under the hood so to speak.

At the end of the article, in the section about MTF, it seems as if the maximum effective resolution of a digital sensor could never exceed the Nyquist Limit of half the total sensor resolution. But if I recall correctly, some of Graeme Nattress' recent work at Red with color science, he has significantly improved on this theoretical limit.

Have I got this wrong? Anybody have information to share on this?

Or maybe I just need to wait until Part II comes out....

 

[David Mullen ASC]

Perhaps it's how you define "maximum effective resolution"?

It's something I've always wondered -- for example, let's say you do a 4K scan of 35mm film and then reduce it to 2K RGB -- in theory, the image should resolve 2K but can a 2K RGB "bucket" (files) actually hold 2K in measurable resolution per color channel without aliasing? Or would Nyquist say that any detail above 1K on a 2K file would be prone to aliasing?

So perhaps you just get into a situation where we normally hold more than half the measurable resolution in a bucket even though that risks aliasing... because real-world subjects don't always hold a fine grid patterns that would make aliasing visible, so it's OK, safe-ish, to aim for storing something like 3/4 of the resolution of the pixel size of the recording format?

In other words, while limiting detail to 2K would be safest for a 4K container, we can risk 3.2K-ish because most real-world detail in the 2K-to-3.2K range is unlikely to cause noticeable aliasing?

 

[Marc Wielage]

It's something I've always wondered -- for example, let's say you do a 4K scan of 35mm film and then reduce it to 2K RGB -- in theory, the image should resolve 2K but can a 2K RGB "bucket" (files) actually hold 2K in measurable resolution per color channel without aliasing? Or would Nyquist say that any detail above 1K on a 2K file would be prone to aliasing?

We found out ten years ago at Cinesite that if you do a 4K film scan and then immediately downrez it to 2K, it winds up looking better than an original 2K scan. Our technical staff christened it "2K+," since the differences were apparent. We used it for lots of major projects before the company shut down in 2004, but the technique is still being used all over LA (and probably NY, Chicago, London, and other major DI and VFX centers).

In other words, while limiting detail to 2K would be safest for a 4K container, we can risk 3.2K-ish because most real-world detail in the 2K-to-3.2K range is unlikely to cause noticeable aliasing?

Aliasing can be controlled to a point, and there are de-aliasing plug-ins that can cut this down to a manageable level, even on a shot-by-shot basis. I had to do this on my HD transfer of League of Their Own some years back, because the grandstand bleachers drove us crazy with aliasing. We had to tweak this again for the downrez DVD release as well. I think we could do it better today, 10 years after that early transfer, since I could window the aliasing tweaks to horizontal-only or vertical-only (which we couldn't do back then). 4K scans are always checked under optimum conditions just to see what the 4K delivery will be like, for the handful of films that require 4K delivery (usually as DPX files).

A lot of this boils down to doing a few tests, looking at it on two or three different monitors (including 2K projection, 1080 HD, and even standard def), just to make sure you're not going to run into "the jitters" later on. Horizontal blinds, radiator grills, and picket fences can be a killer, as can pin-stripe coats. We sweat all this stuff on a shot-by-shot basis in the final color session, and sometimes it's open to endless debate. But it can always be minimized and controlled, given enough time.

 

[Christopher Probst]

I battled for quite a bit juggling just how to cram all of the possible angles on the digital sensor topic into manageable lengths... I originally considered it a three parter. More to your question, there are several intertwined factors all operating simultaneously, so you can't talk about one without considering the other. I will say with the dozens of years of development of more and more sophisticated demosaic processing algorithms, people like Graeme have been able to push this theoretical limit beyond what you should be able to see on paper... And a good thing too!

For me, even my old argument that a 4K Bayer camera can never be considered a true 4K delivery camera is now solved with the Epic's 5.1K sensor... With Red's sophisticated image processing. The Epic IS now a true 4K DELIVERY camera! That should keep us at hay at least for a few years...

Graeme, Jim... I still wish you guys were included in the articles. I hope you now see I have nothing but the best, honest intentions.

;-)