Camera Infrared Filter Resolution and Focus Shift Testing
One of my goals while comparing different infrared lens filters was to perform resolution testing, to get actual MTF50 numbers. People often hand-wave about infrared filters ruining the photo resolution, but they have no numbers to back their claims up. Theoretically, infrared photography should have lower resolution due to its longer-than-visible wavelength of light. Camera sensors mainly respond to the IR light as ‘red’, which is only a fourth of the sensor Bayer “RGGB” pixel population, so that should also lower the resolution.
Cameras with opaque IR filters mounted on their lenses can’t auto-focus, either. You need to focus and compose the shot without the IR filter attached, and then add the filter to take the shot. If you’re lucky, your lens has an IR focus-compensation mark on its focus scale to help you manually shift the focus before taking the shot.
I’m not going to discuss cameras that have been permanently modified to shoot infrared. I’m only talking about mounting different infrared filters onto off-the-shelf digital cameras.
With these three strikes against it, infrared photography is bound to suffer from lower resolution. But how bad is it? I set up my large (size A0) resolution chart to find out.
I chose my ancient Nikkor 105mm f/2.5 pre-AI lens for testing, because it’s as good as I have for handling IR light, and the longer focal length is ideal. Wide-angle lenses are poor subjects for resolution testing, because the required close distance from the chart leads to unrealistic testing. This old 105mm lens can’t be mounted on new cameras that have a “meter-coupling lever” used for setting apertures. For this reason, I used a Nikon D5000 that doesn’t have one of these levers on it. I can’t buy a Nikon kit for this lens to upgrade it to “AI”, which I have done for my Nikkor 20mm f/4 lens, for instance.
After photographing my resolution chart, I discovered that the chart fiducial targets (mainly little black squares) are either semi-transparent or fully transparent to infrared! Shorter IR wavelength filters produced medium-grey target photos, and longer IR wavelength filters made the targets completely invisible! What to do?
My resolution charts are printed from a high-end, large-format inkjet printer that uses dye-based inks. It never occurred to me that IR light would penetrate right through these chemicals. Doh.
I tried printing a focus chart (11” X 17”) using a laser printer next, since those prints are based on toner particles instead of dyes. I don’t have access to laser printers that print larger than 11X17, but I figured I could still get reasonable measurement numbers from my analysis software. The laser printer black toner powder contains “carbon black”, which still looks black in infrared light. The laser print quality isn’t as good as my high-end inkjet chart, but at least my camera can now see the chart.
My next problem was focus. Even using the focus-compensation mark on my lens, the resolution chart photos looked pretty soft. I couldn’t tell if this softness was due to poor focus or the IR filter effects. Maybe both.
My analysis software (MTFMapper) also has the ability to evaluate focus, using a separate “focus” chart that is rotated at 45 degrees about the vertical. This focus chart also has little squares (or trapezoids) that can have their edges measured for resolution.
I decided to perform my analysis using the focus chart instead of the resolution chart, since I could tell where sharpest focus was, and I could still get some resolution measurements. These resolution measurements are only at a few chart locations, instead of across the whole camera sensor that the resolution targets provide. I figured that this was a reasonable tradeoff. This technique made my tests insensitive to the inevitable focus errors.
I tested three different IR filters. The first is the Hoya R72, which is tuned to 720 nanometer light (short-wave IR). The second filter is the Neewer 850nm, which I estimate to be more like 740nm instead of 850nm. I don’t have the necessary instrumentation to measure spectral response, so this is just a guess. The third filter is the BCI 850nm (long-wave IR).
I added the Zomei 850nm filter to the testing. It appears that its spectral response matches the BCI 850nm filter. I believe that the Zomei 850nm is superior to the BCI 850nm in how even the lighting is across the whole filter. The BCI has a slight light-dark variation that looks like concentric rings; the Zomei is completely even.
The focus chart design, showing where the lens should be focused
The picture above shows what the focus chart looks like. The left-hand side of the chart is rotated away from the camera by 45 degrees about the chart center vertical axis. You focus on the right-hand edge of the large middle rectangle (as shown above in red), and you expect that the nearest and farthest little black squares will be out of focus.
Focus Chart, No IR filter, NEF format
The focus chart detail shown above, using an un-sharpened RAW format photo, shows the measurement results without any filter. The manual-focus lens has peak resolution measurements of 0.19 cycles per pixel, or an MTF50 of 39.8 lp/mm. The plane of best focus is a little in front of the large focus-target rectangle. The baseline for this lens is therefore 0.19 c/p resolution for un-sharpened raw photos at this aperture.
The lens was re-focused after taking the above “no filter” shot, to compensate for the IR focus shift. I used the IR focus-shift dot on the focusing scale to manually shift focus, which I then left alone for all subsequent shots with the various IR filters. Nikon hasn’t said what frequency of IR light that this dot is calibrated against (but I bet it’s for short-wave IR).
Infrared focus dot sample. Shift focus by this amount.
Different manufacturers will designate the infrared focus shift with different marks (if they bother to do it at all). Some marks look like little diamonds instead of circles.
Focus Chart Detail, Hoya R72 IR filter, NEF format
The lens has peak resolution measurements of 0.13 cycles per pixel using the Hoya R72 filter, or an MTF50 lp/mm of 27.2. The plane of best focus is right at the leading edge of the large rectangle, where it should be. The resolution has taken quite a dip, going from 39.8 (no filter) to 27.2 lp/mm (Hoya R72).
Without changing focus, I then switched to the Neewer 850nm filter.
Focus Chart Detail, Neewer 850 nm IR filter, NEF format
The Neewer 850nm filter is measuring 0.13 c/p, which is the same resolution as the Hoya R72 filter (MTF50 lp/mm of 27.2). Notice, though, that the plane of focus has shifted farther away from the camera and away from the target rectangle leading edge. Because this filter is about a stop slower than the Hoya R72, I assume the Neewer 850 is blocking the shorter, more energetic IR wavelengths compared to the Hoya R72. The longer wavelengths aren’t focused as well with this lens, and therefore the plane of focus is shifted farther from the camera. I really don’t believe that this Neewer 850 filter blocks wavelengths up through 850nm, though, based upon my other IR filter (BCI) which claims to be an 850nm filter as well.
Focus Chart Detail, BCI 850nm IR filter, TIF format
My MTFMapper software couldn’t analyze the raw-format shots from my BCI 850nm filter. If I converted the photos into TIF, however, the software could analyze them. The focus plane has alarmingly shifted much farther from the camera, with the longer-wave IR light. The lens just can’t bend this IR light enough, so the focus shifts more and more as wavelength increases. The Nikon IR focus-shift dot is definitely not sufficient to work for this filter.
Since I had to convert the photo into TIF format for the sake of the software, it got some undesired extra sharpening by the conversion software. As a result, the resolution peak measurements of 0.17 c/p are bogus. I know that my photos that are converted into other non-raw-formats (jpeg or tif) measure artificially higher resolution.
Focus Chart Detail, BCI 850nm IR filter, NEF format, New Threshold
I discovered I could lower the MTFMapper software threshold setting, and it could then analyze the raw-format photo. The resolution of 0.13 c/p matches the other IR filters (Hoya and Neewer).
Focus Chart Detail, Zomei 850nm IR Filter, raw format
I recently got another filter to analyze: Zomei 850nm. This filter looks very similar to the BCI 850nm filter. The resolution peak of 0.12 c/p is basically the same as other filters, given the measurement tolerances can easily vary by 0.01 c/p. In this photo, I shifted the IR focus compensation by double the amount of the IR mark on the lens. The focus compensation still needs a small amount of additional closer-focus than I gave it.
As expected, all of the IR filters I analyzed decrease lens resolution quite a bit (at least 30%). All of the tested filters essentially match each other with their impact on lens resolution. I was surprised to see that the really inexpensive Neewer 850nm has the same resolution as the Hoya R72; I would have expected it to be worse.
What was really unexpected, however, was the large variation in focus shift according to the wavelength of the IR filters. The infrared focus-shift marks on lenses are very approximate at best. The only consistent theme seems to be that you need to focus closer with infrared compared to visible light. You should stop down the aperture to avoid ruined shots due to the lens being out of focus.
If you do much shooting with a long-wavelength IR filter, I’d recommend that you make your own custom mark on your lens for the correct focus shift. Stopping down the lens may be sufficient to hide the focus shift, if your lens is a wide angle. Beware of stopping down your aperture too much; IR hot spots or concentric rings might start to appear.
The BCI 850 filter, which is dimmer than the Neewer 850 filter by about 3 more stops, causes a huge focus shift. I believe this is because the BCI filter only passes longer IR wavelengths, which the lens can’t focus (bend light rays) as effectively. The BCI 850 is only useful for black and white, because the color information at these wavelengths is mostly eliminated.