Idaho Airships, Inc.

How much more than 12.7 is 16.6? What sort of advantage in print output does the difference translate to?

Canon 5D: 12.7MP, 4,368 x 2,912, demosaiced pixel area 67µ²

Canon 1DsMII: 16.6MP, 4,992×3,328, demosaiced pixel area 52µ²

Consider any one horizontal line of pixels: The 5D will provide 4,368 resolved points at best, while the 1DsMII will provide 4,992. The ratio is .875, meaning that for every 10 1DsMII pixels the 5D will provide only 8.75. Even though there are no partial pixels on the resulting RGB raster, Bayer Array functions will produce pixels reflecting the partial influence from pixel to pixel during demosaicing, so the ratio is appropriate for our consideration. In other words, 1DsMII provides 10 truly discreet pixels for every ~9 that the 5D provides, in a straight line, and that’s presuming we haven’t exceeded our pixels as explained in my previous post.

Now consider our print output. Let’s use a nonstandard enlargement of 36″ with all image elements in perfect focus in order to ensure that even minor advantages have the opportunity to manifest. Your “1440 dpi” 8 ink inkjet printer does not actually produce an image of 1,440 full color composite dots per linear inch, even with sophisticated dot depletion, dot morphology (rumors…), and other algorithms. The “1440″ and “2880″ figures are actually “addresses per inch.” Obviously, this is yet another deeply technical topic best left for another post, but it will help to remember that each ink is binary (it either is or isn’t deposited), that the ink droplets cannot be deposited on top of each other, and that the entire matter relies upon the limits of human visual acuity-that is, the viewer’s inability to distinguish the various color dots-to provide the illusion of continuous tone.

In considering the matter of limited human visual acuity, perhaps the most important issue is that human visual acuity is wildly variable, even given the same test subject. Various pathologies (e.g. diabetes, macular degeneration, presbyopia) play a part, and like a camera, light source characteristics including direction and intensity are also germane. Even one’s emotional state can greatly affect both their physiological and cognitive ability to “see.” For our immediate purposes, it is important to note that humans (presuming 20/20 vision) cannot distinguish detail of less than 3/1000 of an inch (equivalent to about one minute of arc) at a distance of 12 inches (which is generally accepted as a comfortable close-viewing limit). That’s about 1/300 of an inch, which corresponds imperfectly but usably to the actual capabilities of most contemporary professional grade inkjet printers. Trust me on that one, and remember that I’m not talking about the CoC.

A good resources on issues of human visual acuity concerning color with a related exploration of halftoning is this .pdf.

In our “test” 36″ print the 1DsMII is providing an uninterpolated 139 bona fide full color composite dots per inch of resolution. In the same dimension, the 5D is providing 121. This is obviously in the absence of any hostile post processing such as Transforming, Rotating, and so on, that would produce interpolative pollution. At normal viewing distance (~48″ if the image wast taken with the “normal,” or ~50mm lens), the spatial difference between the two is moot. Upon artificially close scrutiny, presuming 12″ to facilitate accommodation (comfort, if you will), the difference of 18 dots per inch, or ~1/8 the functional resolution of the 1DsMII, or ~1/15 the actual bona fide resolution of the properly prepared (RIPped, for this test) print may, at first blush, seem material-but it isn’t.

Detail doesn’t occur across an entire print. Details occur in very limited spaces. An edge, even a slanted one, may exist across only 4 or 5 pixels in our 300dpi print (the camera produces ~5000 dots, the RIP translates them to 10,800, so each print dot represents about 2 camera dots). This means the difference between the two camera resolutions, a ratio of ~10:9, must be effectively distinguished over a space of half of either side of the ratio.

That doesn’t work on a slanted line, and it is even more problematic to discern the difference on a perfectly registered (with the raster grid) edge.

And the matter of human visual acuity? Detail differences between the two camera resolutions within that 4-5/300 of an inch at a distance of 48″ are moot. Even at the artificial viewing distance of 12″ (for our 36″ print) the differences are far beyond the capacity of the ideal human eye to discern.

Obviously, reducing the size of the print does nothing to exacerbate any detail discrepancies relevant to our exploration.

My conclusion: It is the autofocus accuracy of the 1DsMII that may be its greatest advantage over the 5D in terms of the ability to reproduce detail. However, the pixel pitch of the 5D and the lensing limitations discussed previously seem to produce a more promising potential for spatial and tonal advantage from the 5D in the application of handheld aerial imaging.

If I have suffered any errors in logic through this testing, please bring it to my attention. This is not an issue that I encounter on a regular basis-therefore, I recognize that my opinion may not be particularly well developed.

L