As we have seen, we encounter a conundrum in trying to compare the "image quality" (perhaps subjective) of two cameras with images of different pixel dimensions.
Let me suggest, in the abstract, without reference to any of the dogs in this catfight (to mix a metaphor), a way we might deal with that dilemma.
I will use as a model an approach that involves the making of prints. I do not mean to suggest that doing so should be part of our actual "evaluation" protocol; it just serves to illuminate the principles involved. It is in fact an idealistic model, one that we probably couldn't fulfill even if we wanted to.
Suppose we have/do this:
• The two cameras regard the same scene with identical framing. That of course means that the lenses that are involved in the comparison have appropriate focal lengths, which the lenses we wanted to use might not. It also assumes that the two cameras produce images of identical aspect ratio. (If that's not true in the real case, we can fairly readily deal with that. But not right now.)
• We next print the images from the two cameras to the same size. We do this with a printing chain whose resolution capability does not become limiting for either image (that is, has higher pixel dimensions in the print than the pixel dimensions of either source
image). This will of course require, for both prints, mediation between the pixel dimensions of the image and those of the print. We assume this mediation to be done by the "best available" interpolation algorithm.
• We then give the two prints to our panel and have them opine on the image quality (in any senses of that we care to ask about). The viewers may, if they wish, concentrate on a small region, perhaps by using a hand magnifier. But in fact their observations there may be misleading - two images that seem to have the same quality when examined at a "normal" angular size may not reveal the same properties when a small region is magnified. And this in fact is one of the central conundrums of image inspection. ("This one is clearly the best image. Why does this little piece of it look worse under the magnifier?")
Funny, it works that way for models.
Since the ultimate purpose of a camera, in general, is to produce an image in "viewable" form, it would seem as if this fanciful process would be the basis for "meaningful" comparison.
Now it is tempting to say, "but, but, camera X has a larger sensor", or "camera Y has greater bit depth out of the ADC." But this is really like, "I really like Michelle's pudding better, but you have to keep in mind that she mixes hers with a Kitchen Aid K-45, but André mixes his with a K-55".
[Michelle is well-advised. The bowl of a K-55 will not stand up by itself on the countertop, rather a nuisance in the overall scheme of pudding-making.]
Of course, in our context, we can hardly rely on sending around stacks of prints to allow these comparisons to be made by our cadre. So let's see how we can transform this model to a a paperless version.
I'll start by retaining the original presumption of equal framing of the subject.
Then I suggest the following. Take each of the two source images and "uprez" both of them to the same pixel dimensions (greater than for either of them originally). Do this with the "best available" interpolation algorithm.
Then examine them both (in their entirety) on-screen at the same size. This is probably best done with a fairly large screen.
Period.
Now, how about looking at "crops" of each?
Well, if the crops have the same "angular magnification" that we would experience when looking at our prints under "standard" conditions, OK. (This in effect gives us a larger virtual screen than our actual screen.)
But if the "magnification" of the crops is significantly greater than that, we get into trouble. We noted above that differences seen there may not have the same implications as when viewing the whole image. And if our object is to compare "delivered image quality", we can only do this in a "normal viewing" context.
If we are trying to evaluate the quality of two full-sized copies of Michelangelo's "David", as complete pieces of art, we dare not do it with a magnifying glass.
On the other hand, if we are interested in looking for traces of the different effects that a K-45 and K-55 mixer have on the particles of the pudding. . .
Best regards,
Doug
