retroman wrote:For 2K I can only say, you don´t need it, because the emulsion of the film stops at about 700 - 800 lines.
I've scanned Super8 film at 20K (literally speaking) and have yet to find these mythical lines (metaphorically speaking).
The only reason one stops at a certain pixel resolution is that it can become quite impractical beyond such. But that changes as technology improves. If today I scan film at 5K, it is because I no longer use my Commodore 64 for making use of scans. Once upon a time there was supposed to be no point scanning Super8 film beyond SD, if that. And the blame was put on the source. In reality it was the target domain rather than the source domain that determined the limit. But if the target domain were 1K (for example) that doesn't mean there's no point scanning beyond 1K. One should (at least) scan at twice the frequency of the target, a rule derived through something known as Nyquist Limits. But notice how the limit is defined: in terms of the target, not the source. If the target were 2K then the rule would require one scan the source at 4K.
However even the theory of Nyquist Limits is somewhat useless here if one is trying to describe the film itself. It is based on an idealised formatting of information where the information is arrangeable on a regular grid, and which will cause what is known as "aliasing" if not super-sampled (a method of anti-aliasing). It is that which can otherwise appear as a moire pattern (aliasing). But in film the information is not on a regular grid (although what was photographed might be). But there is no moire pattern that can occur on the film. You can't get that shimmer to occur on film except by transferring it from where it might have otherwise occured prior to the film. Aliasing can occur in the relationship between the pre-film domain and the post-film domain, but the film itself (in the middle) will not be a contributor. It will only act as a bridge.
There is a disconnect between the way film encodes a signal and a regular grid encodes a signal. One can't use the way video/digital is described, in terms of lines, to describe the other. The MTF description of film is an attempt to do that but unlike digital versions of such it relies on a subjective observer to decide where the threshold might be deemed (because candidate objective ones for film are completely unbelievable). With digital no such subjective observer is required. Believable limits are defined by the design of the technology. If you understand the digital design the only need for a digital version of an MTF chart, is to explore where a
lower limit might be deemed to subjectively occur. The believable upper limit is already known.
This does not mean there is no limit to film, only that the limit can not be defined or described by a simple threshold. One can only
deem such a threshold. If you look at the MTF for film, it is a curve that continues off into a no-man's land of uncertainty (where graph designers fear to tread). With digital there is no such uncertainty - one can safely erect a stop sign at an exact location, beyond which there will be definitely no point going.
So where does one draw the line with film? Wherever it's the most cost-efficient to do so. If there is no demand for 4K scans of Super8 then one may as well do it at a lower level. One goes with the subjective observation method (which is a very good and robust method I might add). To put it another way it's only when one is able to exploit the information available from a 4K scan that it might become economically useful to scan at 4K. Until then it will just be redundant information clogging up the post-production pipeline. The redundancy is not a function of the film (nothing in a film is redundant). It is a function of the extent to which one does, or does not, exploit the information. And typically (for whatever reason) Super8 is under-exploited.
Now if you do make use of the information theres no need to stop at the Nyquist limit. One can scan at 8K for a 1K target, and if nothing else one will obtain more bits per target pixel. So instead of a 24 bit signal one could have a 48 bit signal instead, or a 96 bit signal (etc.) the very real benefits of which will not be lost on any colour grader.
C
Maybe one day we'll see a unit with an upgraded camera, although those machine vision units certainly aren't cheap so it will probably necessitate a significant price increase.
