Agreed…
I didn’t have access to my computer so I reused an image from a previous post. Here’s maybe a better comparision:
To the left is the raw input image which was color-corrected to my best abilities. Scan resolution was initially 2880 x 2160 px, camera used for scanning was a see3CAM_CU135, lens a Schneider Componon-S 50 mm. Five different exposures were zoomed down to 1440 x 1080 px for exposure fusion and a color-correction. The result is displayed to the left of the above image.
The film stock is original Kodachrome 40 color-reversal stock from 1981 which (for various reasons) had been developed nearly a year after the original exposure of the material happened. That is probably the reason for the strong grain noticable. The camera was a generic-branded model, original manufacturer was Chinon, Japan. The optics of this camera were ok, but definitly not high-grade.
To the right, the result of a spatio-temporal denoising/processing step at 1440 x 1080 px is displayed. For display in this post, the original images were scaled down to a resolution of 960 x 720 px.
Given, such pronounced film grain is only observable in small movie formats. And grain adds certainly to the aesthetics of the footage. But film grain is produced by the film stock used and has no real correlation to the content of the movie (that is: what is actually imaged in the movie). I consider film grain therefore a creative choice - in fact, you could re-add a suitable amount of film grain back to the recovered footage on the right, if you like. Here’s a small example, to illustrate the concept:
On the right, a little bit of the original grain (40%, to be exact) was added back to the restored image. Something similar to this would actually be my preferred output.
Just a side note: there is actually active research in film compression technology doing something quite similar: these guys analyze the original noisy footage, get rid of the noise which in turn enables them to get much better compression ratios, transmit both denoised content and noise information to the consumer side, and reassemble at the consumer side denoised image and noise information for an approximation of the original footage. I bet this will be standard compression technology in a few years.
About 15 years ago I was manager at a company producing barrier-based 3D displays, using mainly 50 inch plasma screens. The barriers in front of the plasma screen were produced by a local PCB company which sold us not the PCB, but the large film sheet they were normally using to expose the PCB. The barrier patterns needed to be very precise for the whole thing to work. After some try and error, we had quite stable results. If you image such a large print with a very good lens onto fine-grained 35 mm material (Kodak Technical Pan comes to my mind), you should get to get a very nice personal calibration target for tests.
Well, some mathematics and measurements of your optical setup might be required to actually get the real resolution scale of your final 35 mm test target, but that is managable.
Easier to use is ready made calibration material. For example, for Super-8, you can still get from Wittner cinetec SMPTE RP-32 test stock.
Also, there are (optical) companies who are producing calibration targets with extreme precision. Here’s an example. These targets are usually exposed on perfectly flat glas substrates and are actually not that expensive (at least some of them).