ilkka_nissila
Offline
• • • •
Upload & Sell: Off
|
 p.6 #1 · Utility of dynamic range beyond a certain point | |
gdanmitchell wrote:
In reality, no sensor provides the Perfect Holy Grail Of Color, and when we include variations in lighting color (directly, reflections, and light coming through things like leaves) color is highly variable. I suppose that if you are working under color temperature controlled lighting all of the time you can minimize these issues, but otherwise…
Given all of those variables, in my experience you can get pretty much any sensor to look the way you want it too if you are proficient at post.
YMMV.
It may well be possible to make two images captured by two different cameras (with substantially different CFAs) look similar enough by adjusting one of the image by eyeballing each image and making corresponding adjustments to one of the images until a subjective match is satisfactory. This I do not deny. But you cannot make a machine or an algorithm which would automatically do this for an arbitrary scene without having the reference image of the same scene from both cameras available, across all conditions, because the images contain different weightings of the information in the spectrum of light of the original scene and this is simply mathematically impossible to have an automatic method which would do this for all images. I would argue that also a human, no matter how qualified and skillful, cannot take an image from one camera and adjust it to match the image produced by another camera without actually seeing the images from both cameras to base the adjustments on, in the general case.
I might make an image with my DJI drone and then ask you to reproduce how that image would have looked with your Canon, or my Nikon, and then we can compare the adjusted image to my Nikon Z8, for example. I could do this for different scenes (some in bright sunlight, others in blue skylight, and yet some in artificial light, say, fluorescent, LED, tungsten) do you really believe you can predict and make a fixed set of adjustments that will in all cases convert images from the DJI drone to look like the Nikon (or your Canon) using a preset that is identical for all scenes, and implement this preset without seeing the image made by the target camera in each scene? This is simply not even theoretically possible, let alone practical.
Now, why does this matter? If the philosophy of the photographer is to make art that subjectively looks like they want, they can perhaps take any camera and produce images that they like, by adjusting the results to their liking. That's fine, if the photographer has the time to do that. However, if the purpose is to reproduce colors of the natural world, without a subjective component, then different cameras produce different results, and your subjective view is irrelevant since different people will make different adjustments to produce something that they like or how they saw the scene, rather than how the scene objectively looked. If you are prepared to work the images to look the way you like, each image in turn, all is fine. However, I would argue that it would be better for documentary purposes to make cameras that objectively reproduce the spectrum of light in the scene without requiring editing but this is technically too difficult, and compromises have to be made to make the camera practical. I think the main issue with making this is that different people see things differently. Another problem is that resolution and signal-to-noise would be compromised. I recall reading that in the early color film days, there were implementations of color film with dozens of different spectral sensitivity dyes and while it worked, it was not practical for commercialization, so we ended up with just three dyes for image capture, though photographic inkjet printers can these days have a dozen dyes or so.
|
