[Leica] Random observations on resolution (long and irrelevant to the craft of being a good photographer)

[hlritter at bex.net (Howard Ritter)]

Poking around with huge degrees of enlargement and up-sampling (but perhaps 
not irrelevantly so for making large prints of landscapes, etc) in PS with 
files from M9, M240, NEX-7,and D800 (not E), I found:

1. The D800?s 36MP FF sensor with the current Nikkor 35/1.4 at f/5.6 
produces conspicuously better detail near the limit than the M240?s 24MP FF 
sensor with the Summilux 35 ASPH at 5.6 does, and the NEX?s 24MP APS-C 
sensor (same pixel size as a 54MP FF sensor) with the kit 18-55 zoom set to 
produce the equivalent of FF 35mm FL produces about the same image 
resolution as the M. This is not the end-all of important sensor 
characteristics, but it can be an important one under some circumstances. 
What this tells me is not only that a 24MP FF sensor does not put modern 
premium prime glass to the test, but also that even inexpensive modern 
kit-zoom glass would not be outclassed by a 54MP FF sensor with regard to 
resolution. This would seem exactly analogous to the role of fine-grain film 
back in the day (anyone remember that stuff?). One wonders what Leica AG 
(and every other manufacturer?s) engineers make of this fact, and whether 
there is a 54MP camera (M540?) or beyond in their minds. Of course, as with 
Microfile film, the part of the "need spectrum? such capability occupies 
would be very small. Still, Microfile had its enthusiasts beyond 
microfilming documents for efficient filing. I?d like to know what pixel 
count (disregarding tradeoffs in noise etc) corresponds to the innate 
resolving power of the best modern glass at center and optimum aperture. 
Given the improvement produced by the ~25% linear increase from 24MP to 36MP 
and the 50% increase to (an effective) 54MP, it?s clearly at least 1.5 
times, and maybe twice, the linear count of a 24MP sensor (i.e., ~50 to 
100MP). And what pixel count corresponds to the best general-use emulsions 
from the Age of Film (K64, Plus-X, etc) in terms of lp/mm? Anyone have a 
reference? These results also make me wonder about the actual utility of the 
new superpremium normal lenses, the 50mm Summicron ASPH and Nikon?s 58mm 
1.4, with current sensors. Maybe they extend the envelope in which they are 
not outmatched by the sensor further from the center and from the optimal 
aperture beyond what lesser lenses do.

2. Doubling the linear number of pixels H and W in PS produces a clearly 
smoother image, with what appears to be better resolution, near the limit. I 
know that in theory this is illusory, as creating new pixels from the 
averages of their parent and neighboring pixels cannot add new information. 
But the appearance of doing so is strong, and I think this is a result of 
the fact that for the most part, natural subjects are not wholly random but 
have fractal dimensions and high degrees of internal correlation: for 
example, linear or continuous features are common, such as areas, edges and 
boundaries, and so on. Such features are not likely to be confined to a few 
pixels but to extend over many. Multiplying pixels as is done in PS can 
create a powerful illusion of making a linear feature seem better defined 
and sharper. If you took a picture of a wall of tiny square, randomly 
colored tiles such that the image of 4 tiles in a square exactly occupied an 
entire pixel, the original file would make the 4 look like 1, with a color 
representing their average (this is a thought experiment, ignoring the fact 
that we deal, Foveon aside, with single-color pixels and Bayer patterns). 
Pixel-doubling would then produce not a faithful depiction of the actual 4 
tiles making up the square, but an illusion of 4 tiles and an artificial 
average color for each of the virtual tiles. But this is a very unnatural 
situation, and in real life, with natural subjects, what appears at any 
given point in an image is likely to closely resemble what appears at the 
points that correspond to the adjacent pixels, so that pixel-doubling does, 
in at least a semi-real sense, have the effect of increasing the visual 
resolution of the image. I think of up-sampling the original file to 
increase the pixel count as ?unmasking? information that was implicitly 
there as a result of the innate characteristics of the physical world.

?howard