Diffraction and Image Quality
Published July 2010
First revision: January 2014
Diffraction: the foe in near IR
Common wisdom (and optics...) states that diffraction is a dangerous enemy when shooting IR because it increases with the wave length. It is always recommended to avoid f/16 and the like because of the negative effects it has on near-IR (from hereon referred to simply as ‘IR’) image quality.
I have decided to carry out some experiments to see whether there is some truth in the above. No, I am not challenging the laws of optics; I am only interested in checking whether the worsening in image quality is indeed more pronounced when working in the IR domain and how much worse it really is when looking at a picture as opposed to measuring it with some instrument.
I carried out a few very simple tests using two cameras and two lenses.
The two cameras are a Canon 1DsMkII and a Canon Rebel XT permanently modified for IR capture. Hence, one full-frame and one APS-C have been used: the former to shoot in the visible domain, the latter to shoot IR.
The two lenses could not be more far apart both in focal length and in quality. The former is a Canon 100/2.8 macro lens that can shoot f/32, the latter is a Samyang 8/3.5 fish-eye for the APS-C format.
The intuition here is to check whether diffraction problems behave differently when moving from a high performance lens to an “entry model.”
Before looking at the experiments, discussing the results and deriving some general guidelines (hopefully), it is important to point out what these tests are not.
- They are not a comparison between the Canon and the Samyang lenses. It would be eminently unfair given the difference in cost but most of all given the difference in focal length, coverage, etc etc.
- It is not a comparison between the two cameras either. I am sure there will be differences, but this has never been the goal in my investigation.
The goal instead is to check how diffraction evolves when moving from small f/ to larger ones in the IR spectrum and in the visible spectrum. If it is indeed worse in IR than in the visible domain, how much worse is it?
Diffraction: the foe in near IR?
We show below the two pictures taken with the 100/2.8 and with the 8/3.5.
The first picture has been taken with the 100/2.8, the second with the 8/3.5. One can notice that the Samyang cannot cover a full frame. The red arrow in the former indicates the church in the background that we will use as a target.
All images have been converted from RAW to TIFF 16bit using Capture One; color space has been ProPhoto. All images have been processed with AutoLevel-AutoContrast-AutoColor in Photoshop. A light unsharp mask has been applied (100, 0.7, 2). Final quality of all images could have been better for sure but I was interested in relative comparisons. Needless to say a tripod and mirror lock with remote trigger have been used.
The first experiment was to shoot three images at f/2.8, f/22, and f/32 with the church at the center of the image. The 100% crop in IR (i.e., with the Rebel XT) and in visible light (i.e., with the 1DsMkII) is shown below.
Hold on! But wasn’t diffraction supposed to get worse at smaller and smaller apertures? Well, yes, but in the real world there is a thing called focus shift that affects IR. Diffraction does get visibly worse from f/22 to f/32, but the small depth of focus at f/2.8 highlights the fact that the lens was not properly focused (of course it was, in visible light...). So, if there is a lesson to be learnt here is that the focus shift is potentially so detrimental to image quality that one is better off accepting some diffraction in exchange of more depth of focus ... to play it safe.
As one can see from the three crops above, the lens was focused correctly in visible light (and the AF assist of the camera gave also its enthusiastic OK). Here we can see the honest and expected behavior: the quality of the image degrades for smaller and smaller apertures.
But how much worse is diffraction when moving from visible light to the IR domain? The crops below can help us get some idea. We compare the IR desaturated crops to the desaturated crops in visible light (to remove any influence of colors). We have not compared the two at f/2.8 for obvious reasons.
All right, optics never lies. Yes, diffraction increases in both cases as apertures get smaller, but the behavior in visible light is substantially better. Indeed diffraction hurts when shooting in the IR domain (modulo the focus shift, though!).
So far we have kept our target in the center of the image. Let’s have a look if the situation changes when moving it toward a corner. The experiment is less obvious than it seems, because my experience with some lenses when shooting IR has been of a bizarre behavior at the corners. That is, a lens that was perfectly ok in the center would exhibit increasing out-of-focus moving away from the center to reach a massive focus shift at the corners. My (amateurish) guess was that the combination of longer paths that the rays had to travel, combined with some optical structure, combined of course with the fact that the designers had not optimized the lens in near-IR to start with created the monster. I say ‘monster’ because these lenses were completely unusable.
So, let’s have a look at the corner. First the Rebel, then the 1DsMkII.
Besides a clear worsening of the image quality when shooting IR (note how good the 100/2.8 is in the visible domain), similar considerations apply as before: diffraction worsens with smaller apertures as to be expected, the focus shift in the IR domain calls for staying away from shooting at full aperture or nearby, thus accepting some degradation because of diffraction in exchange of (the absolutely necessary) depth of focus to compensate for the focus shift. And again diffraction is worse in the IR domain, without being a fanatic pixel peeper (see the two crops at f/32, for instance).
To conclude, let’s try now the Samyang fish-eye on both cameras. The target here has been changed to include at least three planes of focus (2m, 20m, and infinity). Once again all 100% crops have been desaturated to eliminate the visual influence of colors.
The humble Samyang turns out to be an honest performer, actually a rather surprising one. It is very interesting (and reassuring...) to notice that in this case, where focusing was not an issue, the image quality in the IR domain at f/3.5 is clearly better that at f/16 (look at the tiles on the roof, for instance). Urrah!
One last comment has to do with live-view and focus peaking. The cameras used in these experiments were second generation DSLRs. While the laws of optics have not changed in the meantine, notable advances in technology have happened that can change some of the considerations made above. The most important one is the availability of live-view and most of all of focus peaking. With focus peaking one can utilize a digital camera permanently modified for IR photography without having to worry any more about near-IR focus shift. That is, as long as the photographer does not want to use autofocus. When picture taking is relaxed, the camera is on a tripod, manual focus can be used, focus peaking is a game changer, no question about it.
In the latter case diffraction does become the foe to stay clear of, as poor focusing no longer represents a present danger in near-IR photography when focus peaking is used. In all other cases, i.e., when you really want to shoot in near-IR that hummingbird in your backyard, better keep your camera f-stop relatively … overweight, and diffraction be darned.
(c) 2014 Marco Annaratone, all rights reserved