Olympus E-PM2 for multi-spectral imaging  

Note - The camera described on this page, serial number BGC505015, was stolen from me in 2016, in connection with moving to another address.

This page discusses an Olympus E-PM2 Micro 4/3 camera converted to multi-spectral imaging by eBay seller eeassa. This seller offers a range of digital cameras converted for NUV, NIR and multi-spectral photography. The last category of cameras are converted by replacing their built-in UV-cut, IR-cut and anti-aliasing filters with Spectrosil windows. This seller directly provides the camera (a manufacturer-refurbished specimen purchased by himself), which is a significant advantage for users not based in the US. There is no need to buy a camera yourself and ship it to the converter, as required by other US-based conversion services. This avoids the payment of two-way import duties - first into the US, then back to your country - and half of the associated shipment fees and risks involved in international shipments.

Spectrosil is a synthetic fused silica material (see here (dead link) and here) that guarantees the transmission of UV radiation down into the UV-C range. This material is also free from fluorescence in the usable spectrum, and from optical defects like bubbles. The actual spectrum recorded by the converted cameras does not extend much below 300-310 nm, because of limitations of the sensors and/or sensor-package windows and coatings. The package window cannot be removed or replaced without compromising the durability and cleanliness of the sensor chip. Nonetheless, it does not hurt to use a Spectrosil replacement window, even if its transmission spectrum largely exceeds the sensitivity limits of the camera sensor. In this way, we do know that the replacement window, unlike poorer replacement windows sometimes used by unscrupulous conversion companies, is not a limiting factor in the performance of the converted camera.

It is possible to simply remove the original filter from a mirrorless camera, instead of replacing it with a transparent window. Doing this, however, no longer allows the lenses designed for this camera to focus at infinity. Usually, they only work in the close-up range or, at most, are good for portraits. I did this type of conversion on a Panasonic Lumix G3, and had to modify a few Micro 4/3 adapters and lenses to restore their infinity focus with this converted camera. This works, but in the end it is expensive and time-consuming to perform these modifications, and this was one of the reasons that convinced me to purchase this converted E-PM2. Unlike the G3, the E-PM2 offers sensor-based image stabilization, which is another advantage in hand-held photography. Additional reasons were that this seller was warmly recommended to me by another UV photographer, and that his prices are reasonable. In fact, I would have spent about the same amount by purchasing a new, non-converted E-PM2 from one of the large discount retailers in Sweden. This country is not exactly known for bargain prices in any context, and no doubt I could find a cheaper E-PM2 on eBay or purchase it in another country during one of my trips.

General E-PM2 characteristics

The E-PM2 is a small Micro 4/3 camera, even compared to other Micro 4/3 cameras like the E-M5 and E-M1. Compared to my first multispectral camera (converted Nikon D70s), the E-PM2 is really small and lightweight (Figure 1). The lack of an eye-level viewfinder is partly responsible for this. An optional viewfinder can be mounted on the accessory and flash hotshoe, but doing this prevents the use of a flash. There is no built-in flash. A miniature, dedicated Olympus flash can be mounted on the accessory port and hotshoe, and it can also be used in master mode, controlling up to four groups of external flash units. The rear LCD screen is large, but can neither tilt nor swing. The grip for the fingers of the right hand is obviously too small for a normal-sized male hand, but the camera can still be held safely, helped in this by the small grip for the right thumb at the rear of the camera.

So far, I found it impractical to use an Arca-compatible plate for attaching this camera to a tripod or macro stand. Most Arca-compatible plates do fit (although they are much too large compared to the camera), but all those I tried block the battery and SD card door. Things are made even more difficult by the 1/4" tripod socket of the E-PM2 being offset toward this door, rather than sitting right under the lens mount (Figure 2, left). This forces the removal of the Arca plate every time it is necessary to open the door. Apparently, no custom Arca plates or L-shaped brackets are available for this camera model. A partial solution to this problem is using a lens adapter with built-in tripod shoe, to which an Arca-compatible plate can be attached. If this is not possible, a large Arca plate can be attached at the bottom of this camera, and removed whenever the card or battery need accessing. I am considering cutting out part of such a plate to allow the door to open without removing the plate, but there is a possibility that such a modified plate will not provide a sufficiently stable attachment to the camera.

Controls of this camera are not numerous, and many simple configuration changes require a deep-dive into the LCD menus. Nonetheless, in addition to the essential buttons there is a rotating dial (which also serves as the standard set of four-way arrow buttons) and a programmable function button is located at the right of the shutter release (Figure 2, right). I use the programmable function button to magnify the live view image for manual focusing, which is the control I use most frequently with manual-focus lenses.

The camera is turned on and off by pressing a small flush button at the left of the shutter release. A blue LED near this button indicates the power status.

In spite of the problems caused by the small size and limited number of controls, I love this camera for its light weight and portability. It allows me to carry a multispectral camera and a small UV-capable lens like the G.Zuiko 20 mm f/3.5 without noticeably increasing the weight of my backpack. The weight of the camera body with battery is 269 g. The battery is type BLS-5, which is smaller and lighter than the BLN-1 used in the E-M5 and E-M1. This means carrying around one more spare battery on a longer outing, and an extra charger on a trip. Luckily, genuine Olympus BLS-5 batteries are much cheaper than the overpriced BLN-1 batteries.

Compared to my multispectral Panasonic G3, the E-PM2 offers the important advantage of sensor-based image stabilization. With the 20 mm lens shown above, I can reliably shoot hand-held at 1/4 s, or even longer exposures if I shoot a few images and choose the best one. Of course, image stabilization is useless with moving subjects, like windswept flowers.

The conversion

I do not have details on how the conversion of this camera was actually carried out, and in particular what precautions were taken to prevent, as much as possible, the contamination of the sensor by dust while replacing the window mounted in front of the sensor package. Established companies like LifePixel convert cameras in industrial clean-rooms, but even this cannot guarantee a complete absence of dust trapped between the sensor and the replacement window, especially if the camera to be converted has been used or its original factory packaging has been opened.

Dust trapped between sensor and replacement window cannot be removed without disassembling the camera and separating the replacement window from the sensor. This procedure is likely to cause new dust particles to be trapped between these parts. The original NIR, NUV and anti-aliasing filter is mounted on the sensor in a clean-room environment, and its cleanliness is assured by the fact that none of the involved components has left industrially clean environments before this assembly phase. Camera producers generally do not replace the built-in filter even if damaged, because of the impossibility of preventing dust contamination. They prefer to replace the entire filter and sensor assembly with a new pre-assembled one, because this is the only practical way to guarantee that there will be no dust contamination.

The replacement window of this E-PM2 is apparently sealed around its edge with silicone, without any silicone contaminating the surface of the window (which would be practically impossible to clean out). The job seems to be well done along most of the window edges and corners (Figure 3, top), except for one edge that looks clumsier (Figure 3, bottom). A large chip of glass seems to have broken off from the edge of the window here, during its cutting and grinding. Still, also this region seems to be well sealed against contaminating dust, and the chipping is well outside the image-collecting area of the sensor. The figure also seems to show strips of double-adhesive tape between the sensor and replacement window. They were applied in separate pieces (as opposed to the single-piece, seamless "frame" normally used in the camera industry) and leave gaps near some of the corners (Figure 3, top). I don't know whether this tape was originally holding the built-in filter, or was added during the camera conversion. Since the silicone sealant fills these gaps, the latter do not seem to constitute a problem.

Despite the slight irregularities discussed above, the window seems to form a good seal against the sensor package, and I did not detect any obvious, non-removable dust particles in images recorded by this converted E-PM2. Perhaps I might have received an esthetically better conversion through one of the major conversion services like LifePixel, but at an additional cost of perhaps 200-300$, and without getting a correspondingly improved functionality.

Like the original filter, the replacement window is directly attached to the sensor package and moves together with the sensor when image stabilization is active. Switching off this E-PM2 causes the image stabilization mechanism to gradually shift the sensor and its replacement window to the right and bottom, then back to its center position. This is accompanied by a buzzing noise. The E-PM2 has the piezoelectric "dust shaker" found in other Olympus models, which is more effective than the sensor-shifting shaker used in other cameras. This dust-shaker is removed during conversion. Nonetheless, the sensor-shaking at power-off might help to remove some sensor dust when the camera is switched off.

Custom white balance

Setting an appropriate white balance (WB) in NUV photography is an arbitrary process, since any UV image with any WB necessarily records false color. Since I often take comparison shots of the same subject in VIS and with different NUV-pass filters, I decided years ago to set a custom in-camera WB that produces reasonably correct color rendering in VIS images recorded with a filter that cuts NUV and NIR. This type of filter is necessary to approximately restore the color faithfulness compromised by removing the original built-in filter. The custom WB then further compensates for the different transmission spectra of this filter and the original built-in filter. For these tests, I used a B+W 486 UV- and IR-cut filter, which has a cyan color not unlike that often seen in built-in filters.

I subsequently use this custom WB for all multispectral imaging, including NUV, VIS and NIR. This of course produces plenty of false color in NUV images, depending on the filter used. When the false color is distracting, I convert the images to BW by desaturating the image (i.e., equalizing all three color channels).

The above image (left) shows a Color Checker Passport with the custom WB and the UV-enabled electronic flash (Bowens 1500Pro) used to create the custom WB. This is the custom WB automatically produced by the E-PM2, without manual correction. Gray tones contain slightly more red than green and blue, which is responsible for the warm "Olympus" rendering. I had to underexpose in order to produce an approximately correct rendering of the red and blue squares (second row, third and first from the right, respectively). This causes the white square (first row, first from the right) to be recorded as gray. A gamma and/or contrast adjustment in post-processing can restore a reasonable "whiteness" if desired.

Above, right, shows the imaging of a PTFE reference target with this WB. PTFE was used as a target because it uniformly reflects in the VIS and NUV, and therefore it is as close as practical to an easily acquired white/gray reference material usable in both bands. For more correct result, a Spectralon target can be used. This PTFE-based material material is more reflective, but also more expensive and more easily contaminated in ways that may alter its spectral response. Since the flash head illuminated the subject area obliquely from its left, the right side of the test image is slightly darker.

NUV imaging

With the E-PM2 custom WB discussed above, a white PTFE target imaged with the Baader U filter gives a red false color (above figure, top row, left column). The same target with Asahi Spectra XRR0340 produces a yellow-orange false color (above figure, center row, left column). The test image with Asahi Spectra filter requires a 2-stop increase in exposure because of the shorter wavelengths. In both cases, this is a "warmer" NUV false color than I am used to with a converted Panasonic Lumix G3, and more similar to a converted Nikon D70s. The Omega 325BP10 filter transmits a 10 nm band centered at 325 nm. It yields a yellowish-green false color with the E-PM2 (above figure, bottom row, left column), and requires a 7-stop increase in exposure with respect to the Baader U. It produces a purer green with the Panasonic G3. The Omega 325BP10 filter transmits a 10 nm band centered at 325 nm. It yields a yellowish-green false color with the E-PM2, and requires a 7-stop increase in exposure with respect to the Baader U. It produces a purer green with the Panasonic G3.

The center and right columns provide a side-by-side comparison of the E-PM2 with Panasonic G3 and Nikon D70s, all converted for multispectral photography, with the three filters discussed above. The test pictures for all cameras were shot at f/5.6, ISO 200, and, for all three cameras, at a flash intensity that gives a satisfactory exposure with the E-PM2. All cameras used an in-camera generated custom WB as described above. This allows a comparison of the relative sensitivity and false colors of the cameras with the three NUV-pass filters I use most often.

The false-color NUV palette in straight-out-of-camera images is obviously different for the three cameras - much more so, in fact, than the VIS palette in test images of Color Checker Passport shot in VIS with B+W 486 filter. The G3 seems to have a slightly higher sensitivity at short NUV wavelengths, but the difference is apparently not in excess of one stop, and at least in part may be an artifact caused by the "colder" automatic custom WB of this camera, relative to the E-PM2. The D70s NUV false-color palette, on the other hand, contains large amounts of red with all three NUV-pass filters. In spite of this, D70s images can be post-processed to reduce the red level and yield images not too different from those produced by the two other cameras. In particular, the D70s records increasing amounts of green false-color at decreasing NUV wavelengths, just like the two other cameras. This is easily seen in D70s images after substantially lowering the red channel in post-processing.

Multiple causes are likely to be involved for the observed differences among cameras, for example, among software/firmware causes:

• The way custom WB is automatically set in each camera, and applied in-camera to recorded images
• The effects of demosaicking algorithms
• The general warmer or colder, more or less saturated, brighter or dimmer color rendering preferred by each camera manufacturer

Even comparing raw images cannot exclude firmware differences, because raw images are pre-processed and encoded in different ways that may affect their color balance. After all, if raw images really contained just unaltered data as read from the sensor, we would not need updated raw conversion software each time a new camera model comes out. Furthermore, it cannot be excluded that also hardware differences are involved, for example:

• The transmission spectra of Bayer color filters may differ, assuming that pigments are proprietary and not identical across sensor makers and types. Differences in transmission are especially likely in the NUV, where Bayer filters don't need to behave consistently because they are only meant for VIS recording.
• Different photosite structures (e.g., CMOS versus CCD, front- versus rear-illuminated sensors). Since different wavelengths are captured att different depths within silicon, the depth of a photodiode within the chip silicon substrate may also be a factor.

On the other hand, there are no major differences in sensitivity to NUV in each band among the three cameras, once false color is removed (e.g. by desaturation). If separate images are recorded with each of the three filters and subsequently reassembled as the three color channels in a false-color composite image, results would not be too different among these cameras.

NIR imaging

The E-PM2 performs well in NIR imaging up to about 1,100 nm. The performance of E-PM2, G3 and D70s does not differ substantially in this respect. Any of these cameras will provide good results for this application.

Summary

The conversion of the Olympus E-PM2 discussed on this page is not done to the standards of some of the companies specialized in these types of conversion. However, the performance of the camera is unaffected by the slightly amateurish conversion and the camera works very well in multi-spectral photography Additionally, the price of the converted camera is reasonable. A further advantage for buyers not based in the US is that the camera is provided by the seller, and you are not required to purchase a camera and then ship it to the seller for conversion.

The converted E-PM2 was compared with converted Panasonic G3 and Nikon D70s, all using a custom WB that produces similar images in VIS. Their straight-out-of-the-camera images display much greater differences in NUV false color than in VIS color.