Low-pass NIR filters

The last few years I didn't do much NIR imaging, but I decided that maybe I should. Digital imaging in the near-infrared with a converted camera is much easier than imaging in the near-UV, which explains, at least in part, why digital NIR images are more frequent than NUV images on the web by perhaps a couple orders of magnitude.

Digital NIR imaging initially focused on producing true and mostly monochromatic NIR-only images, but the emphasis seems to have shifted toward false-color images that mix different NIR wavelengths with varying amounts of VIS by using low-pass filters with a broad cutoff region that extends to varying amounts and wavelengths of VIS light. In some cases, dual-bandpass or triple-bandpass filters have been used (initially for scientific purposes, but subsequently for artistic imaging). Swapping color channels or performing more complex color manipulations in post-processing is also increasingly popular. Al though I may occasionally enjoy looking at these false-color images, I am not particularly interested in re-creating them, because I feel that they are getting father and farther from what the camera actually "sees".

Some NIR photographers seem to like dual-, triple- or quadruple-bandpass filters that transmit in two, three or four narrow wavelength bands, including one in NIR, widely separated from each other. I cannot say much about these filters, except that they can be remarkably expensive and produce "artificial"-looking types of color rendering because their discontinuous transmission spectra make them blind to multiple ranges of wavelengths. They were developed for scientific, security and technical applications, and subsequently re-purposed for artistic/aesthetic photography.

I accumulated a small collection of NIR-pass filters over the years, and recently purchased about half a dozen to use with some of my Olympus and OM System lenses. These lenses (especially the 12-40 mm f/2.8 Pro and 40-150 f/4 Pro) cover most of the range of focal lengths I am likely to need, are reported to perform well in NIR imaging, and accept 62 mm filters. I have been successful with 52 mm NIR-pass filters on a step-down adapter on these lenses, with no vignetting except in the wideangle range (which disappears by stopping down the lens by one or two stops). However, NIR filters are quite inexpensive, compared e.g. to UV-pass filters, and readily available in large sizes.

The Olympus 7-14 mm f/2.8 Pro could be interesting to test in NIR, and could potentially produce unique NIR images, but does not accept front-mounted filters. One or two third-party, expensive filter holders are available for this lens, but only accept a very large filter (100 x 100 mm). It is also very likely that a front-mounted glass filter will not work well with this lens, because of its very high angle of view. I might try a thin rear-mounted NIR-pass gelatin filter instead.

This page discusses several of the filters I collected during the years. In order to make comparisons among filters in a reasonably accurate way, I generated transmission spectra of these filters, produced with the method described here. I performed no post-processing smoothening or noise reduction on the spectra displayed on this page. It can easily be seen how noise is higher near both ends of the displayed range of wavelengths. This is simply a consequence of the lower emission level in these regions of the light source used for this investigation.

The tested filters

In my experience, the large majority of low-cost, no-brand NIR low-pass filters are satisfactory for qualitative NIR imaging. A possible problem is that these filters, and the materials used in the filter substrate, are likely manufactured by a number of different China-based companies. Therefore, it cannot be excluded that their optical properties differ among filters specified as equivalent to each other, and/or will change from batch to batch. It is a well-known fact that even companies producing their own filter glass, like Hoya, cannot guarantee that the optical properties of some of their filters will remain exactly the same from batch to batch. Therefore, my findings should only be regarded as indicative.

NIR-pass filters are typically identified by retail sellers only with a single value of their cutoff wavelength. The most common values are 650, 680, 700, 720, 750, 820, 850, 950 and 1,000 nm. Normally, the cutoff wavelength of a filter is measured at 50% transmission. In itself, this may mean slightly different things, e.g. 50% actual transmission, or 50% of the peak transmission of the filter (which is always less than 100 %). Figure 1 shows the actual transmission.

The transmission spectra in Figure 1 display only the wavelength interval between 400 and 1,000 nm, which includes VIS, as well as the portion of NIR that can be recorded with converted consumer cameras. The theoretically maximum filter transmission (in practice, no filter at all in the optical path of the spectrometer) is indicated as 1 on the y axis. Thus, for example, a 0.8 transmission is equivalent to 80 % of this theoretical maximum.

In the measured range of wavelengths, none of the tested filters displays a peak transmission higher than approximately 0.95, or 95%. The legacy Nikon O65, an orange VIS filter, has the highest peak transmission of the lot, as well as the highest transmission of the lot in the range between 550 and 1,000 nm.

Typical NIR filters with nominal cutoff wavelengths up to 750 nm do transmit visually detectable amounts of red light. 650 nm filters are visually a rich red, and only filters with a cutoff wavelength of 850 nm and higher can be expected to be visually opaque. In addition, red, orange, yellow, green and even blue generic color filters are used in digital "NIR" imaging, because they commonly transmit NIR in addition to varying portions of the VIS spectrum.

In addition to the conventional NIR low-pass filters discussed above, I tested a variable NIR-pass filter, separately discussed in detail here. Two polarizers are among the optical components of this variable filter. For this reason, as general information, the above figure also contains spectra of a circular polarizer and a linear polarizer:

• Kenko CPL (circular polarizer)
• linear pol (Kenko PL linear polarizer)

Among the noticeable things in the above figure, the two orange filters (Nikon O56 and no-brand orange) are a bit different from each other, albeit not substantially so. Unless you really want a specific "look" and even a very slight difference matters, either one would do. The two 850 nm NIR-pass filters also differ, with one being more properly an 820 nm filter. Other than this, the nominal cutoff wavelengths of these NIR-pass filters are quite believable.

It can be noted that NIR-pass filters with cutoff wavelengths above roughly 900 nm display a longer, more gently inclined transmission slope, while those rated at lower cutoffs display a sharper inflexion. Perhaps the filters rated below and above 900 nm, respectively, are made with different optical materials.

Legacy filters tend to be relatively thick, while their modern counterparts often, but not always, are thinner. Thinner filters may work better with extreme wideangle lenses. Increasingly often, modern filters are mounted in thinner filter rings, slightly less likely to vignette with wideangle lenses. Sometimes these filters with thin mounts are sold as premium items, but increasingly one gets filters with this type of mount even when nothing is said about it in the seller's ads.

Conclusions

No-brand NIR-pass filters generally work well and their specifications are reasonably accurate. Small differences can be expected among different makers and batches, just like with VIS filters. Filters with cutoff wavelength above 900 nm seem to have a gentler cutoff slope, while filters with lower cutoff wavelengths have sharper ones.