Live ND on the Olympus OM-1

The Olympus E-M1 X introduced a Live ND setting, which has been inherited by subsequent cameras in the E-M1, E-M5 and OM-1 series. This page only discusses this feature in the OM-1, but much of the discussion should apply also to the other Olympus and OM System models mentioned above.

On the OM-1, in either S or M modes, Live ND can be switched On or Off in Camera 2, tab 1. Computational Modes. In other modes, notably including A, the Live ND setting is disabled.

With current camera firmware, Live ND can be set to a value from ND2(1EV) to ND64(6EV) in intervals of 1 EV. LV (Live View) Simulation can also be switched On or Off, to provide a simulated, approximate preview of the ND effect before shooting.

Live ND implementation

In Olympus/OM System cameras, Live ND records multiple images and merges them together with an averaging algorithm, thus simulating a longer exposure. Unlike focus stacking, which selects different portions of each image, Live ND averages each pixel with the same x and y coordinates among all the recorded images. While shooting hand-held in Live ND mode, the camera also uses image stabilization to compensate for camera motion, so, with some luck, you may obtain good results even when shooting hand-held at ND16 to ND64.

Only the ND image is recorded to the memory card, not the original sequence of shots before applying the ND algorithm.

In some of my tests, the exposure time reported in the EXIF metadata of a Live ND image seems to refer to the actual exposure time of individual shots before the ND processing, not to the total exposure time. In other tests, the reported time seems to correspond neither to a single shot, nor to the whole sequence (see below).

Live ND allows a maximum ISO value of 800. I guess OM System has technical reasons for this limitation.

I have wondered, but don't actually know for a fact, what happens if the illumination conditions drastically change while shooting a Live ND sequence. When shooting in S mode like I did for the test discussed below, would the lens aperture change during the sequence? And in this case, how would it be reported in the EXIF? I tend to think, however, that it would make better sense for the camera to set the exposure parameters before the start of the sequence and keep these parameters constant during the sequence, rather than metering and changing them before each individual shot of the sequence. Metering before each shot would obviously extend the length of time required to shoot the sequence.

Live ND as replacement for physical ND filter

The most common, and most obvious, use for a physical or digitally simulated ND filter is to prolong the exposure time by a a factor of a number of stops. This produces the visual effect, for instance, of waves or running water turning into a wispy fog (or mush, depending on whether you like or dislike the effect). Personally, I find this visual effect overrated and over-used. During the early time of film photography, low-sensitivity film plates and dim objectives forced long exposures in landscape photography (as a necessity, rather than an artistic tool), and this is how the public became aware of the mush effect. I perceive this effect as unnatural when applied to waves and waterfalls, because this is not how we see flowing water, unless its movement is extremely fast.

Live ND can also be used to make moving vehicles and pedestrians disappear from urban landscapes if a sufficiently high ND factor is used (they turn into visible wispy streaks if the ND factor is insufficient), and this can be a way to produce images of crowded tourist spots that appear largely deserted (except for a few people sitting immobile on benches). A give-away of this technique is that sitting people who move their heads during the exposure display ethereal blobs in place of their heads. Another common side effect is that stationary people who move away during the exposure are turned into semi-transparent ghosts (an effect often used in "spiritual photography" in the 19th and early 20th century, which purported to show the souls of deceased people). A digital ND64 may not be sufficiently strong for this use. Head- and tail-lights of vehicles turn into long light streaks in ND multi-second exposures of night-time urban landscapes, another well-known, and in my opinion overrated, effect if done intentionally.

Other uses for Live ND

There is another, not immediately obvious use for Live ND as implemented on the OM-1. This use may or may not apply to the implementation of electronic ND filters in other cameras. It definitely does not apply to physical ND filters inserted into the light path.

In Live ND mode, each image in the sequence is properly exposed, and the ND effect arises from the total exposure time of the whole sequence. Thus, for example, ND64 shoots a sequence of 64 properly exposed images, with a total exposure time 64 times longer than the exposure time of a single shot. An ND64 physical filter, instead, results in a single exposure 64 times longer than without the filter.

This is where things get interesting. The analog sensors of digital cameras suffer (mainly) from two types of noise. One is visible in images shot at high ISO (and, usually, short exposure times). The second type of noise becomes stronger during a long exposure. The results of the two types of noise are comparable, but the causes are different. High-ISO noise originates from the fact that a solid-state sensor has just a single, unchangeable ISO sensitivity, usually called native ISO. Higher ISO values than the native sensitivity are simulated by intentionally underexposing, then adjusting the image brightness, contrast and saturation in firmware after the exposure to compensate for the (often massive) underexposure. Noise is likewise enhanced together with the rest of the underexposed image, and its removal by the firmware causes further losses in image quality.

Long-exposure noise is instead largely caused by drifts in the low-light sensitivity and charge-leakage of individual sensels during long exposure times, and gets worse as the exposure time increases. Most solid-state sensors display a low amount of this type of noise up to exposure times around 1/125 to 1/30 s. The amount of noise rapidly increases above this exposure time.

A part of the long-exposure noise repeats from exposure to exposure (or at least, remains similar in two exposures taken in rapid succession). For decades, digital cameras could be set to compensate by shooting a dark frame immediately after the actual exposure, and subtracting its noise from the first exposure. The cameras could be configured to so automatically, whenever the exposure time exceeded a preset threshold. This reduced the amount of processing of images shot in good light, which did not actually require this type of noise compensation. A few years ago, sensors capable of isolating the dark current during the actual exposure, and suppressing this current from the recorded data, became the norm, and separately recording a dark frame has no longer been necessary.

Another portion of long-exposure noise does not repeat from frame to frame, and is essentially random and unpredictable. This is where the multiple exposures come into play. Averaging these multiple images strongly reduces the amount of this type of random noise. This technique has long been applied in digital astronomy (and film astronomy before that), and Live ND does exactly the same thing.

As a result, the product of Live ND is visually judged by many viewers to be a cleaner, lower-noise image, compared with a single-exposure image of the same subject. The difference is especially detectable in low-light images, while Live ND is generally not worth using with well-illuminated subjects like a sunny landscape (unless you are intentionally after the "mush" effect). Live ND also produces images with a subjectively higher contrast and dynamic range. The latter is a result of averaging, which produces a potentially larger number of individual shades if the math is programmed the right way in firmware. The visual end result has been compared to HD imaging, which is based on a different principle and process, but also affects the recorded dynamic range by optimizing the tonal range used by a specific image without increasing the maximum number of tones that can be stored in the image data (in JPGs, usually 8 bits per color channel, or 2⁸ tones per color channel).

Live ND versus AI noise reduction

Live ND reduces image noise at its source, and is completely different from noise removal in post-processing, which is often performed with AI-derived software. The latter type of noise removal is done by removing the part of pixel-level image detail that the algorithm identifies as likely to be noise. Depending on how well a specific AI-based algorithm succeeds with a specific image, it may mistakenly remove only a minor, or a significant, amount of real detail from the original image, in addition to a variable amount of noise. It may also leave tell-tale signs of the noise-removal process (e.g., a higher amount of noise may remain in specific areas of the processed image, without a justifiable reason for this differential amount of noise in a genuine image).

Nonetheless, there may be good reasons for using AI noise removal in post-processing. When used sparingly, it often succeeds in producing a more agreeable image quality. Additionally, Live ND, because of its multi-exposure nature, is not suitable for subjects that move at a moderate to high speed. The judgement and expertise of the photographer remains an indispensable factor in all parts of the process, from choice of subject and camera settings to post-processing and selection of images for publication.

Live ND noise reduction in practice

The following series of test images was taken with different camera settings, some of them indicated in the captions. The first imagewas taken in A mode and is therefore a single-shot image, the rest are Live ND images. In all cases, silent (i.e., fully electronic) shutter was used. Unavoidably because of the natural illumination and variable cloud cover, the illumination changed slightly during the photography session. The subject was shot from a distance of approximately 30 m with Olympus 300 mm f/4 Pro and MC14 teleconverter (total real focal length 420 mm, equivalent in field of view to an 840 mm focal length on full-frame). Auto ISO was used for all images, with a lower limit of ISO 200 and an upper limit of ISO 800. Note that Live ND does not allow a higher limit than ISO 800. Noise Filter was set to Standard and Noise Reduction to Auto. The lens was mounted on a solid tripod.

The first series shows reduced versions of the complete test images. The rendering of the white flowers on the bright green backgound of leaves looks more natural in the first two images, a little "forced" and unnatural (but still acceptable) in the last two (1/8 s and 1/2 s exposure, respectively). Overall, however, color and contrast remain fairly constant throughout the series. A side effect observable in this sequence is that DOF increases (naturally, this has nothing to do with Live ND) as the aperture is stopped down.

The second image series (below) shows 1:1 pixel crops of approximately the same area, on the table top below the flower vase partly visible at top right. The reflection of a curtain on the table top (light-gray vertical stripe) can be seen moving with the wind from image to image. The dust on the table top is reasonably sharp at f/5.6 (especially considering the large distance between lens and subject), slightly blurred by diffraction at f/13 to f/19 but not really bad, considering the Micro 4/3 format is diffraction-limited beyond f/5.6 to f/8. Many of the individually visible dust particles are sub-millimeter in size. The table top is a glass plate overlaid on a coarse, black woven plastic.

Virtually no long-exposure noise is visible on the relatively featureless, much less dusty surface of the flower pot, in spite of shooting at up to ISO 800 and 1/2 s.

The above test gives inconclusive results, because noise is low in all images. This is not fully surprising, since the scene was illuminated by direct summer sunlight through a slight cloud cover, and the illumination level was therefore reasonably high.

The following subject was shot in the studio with no artificial lighting, only daylight passing through an almost closed lightproof curtain and further diffused by dark semi-transparent cloth curtains. For this test I used the OM System 40-150 mm f/4 Pro at 150 mm.

And 1:1 pixel crops of the three preceding figures:

As expected, ISO 16,000 shows abundant high-ISO noise. The noise is much lower at 800 ISO. However, in the latter image, the long exposure time (4 s) causes several visible hot pixels, in spite of running pixel mapping on the OM-1 just before this test. Apparently, pixel mapping is optimized for shorter exposure times and cannot completely correct the hot pixels at multi-second exposure times. Virtually no hot pixels are visible in the 1/8 s exposure, in spite of the much higher ISO.

Much the same hot pixels are visible also in the ND64 image, albeit less bright. On the other hand, this image shows far less noise in the gray areas, while the amount of detail (e.g. the smaller dust particles along the upper edge of the black slot) are rendered virtually as well as in the ISO 800 single shot. They are invisible, instead, in the ISO 16,000 single shot, probably obscured by a strong noise removal performed in firmware (but still failing to completely eliminate the large amount of high-ISO noise).

As a whole, the ND64 image combines the lowest noise and the best detail. This confirms that Live ND can indeed be used to reduce noise in long exposures of a stationary subject, much like the corresponding multi-image averaging technique routinely used in photoastronomy.

One thing I did not expect is the 60 s exposure time reported by the EXIF metadata of the ND64 shot. This is too long to be the exposure time of a single exposure, and also seems to be shorter than the total time taken by the camera to shoot the whole sequence.

Summary

The Live ND functionality in the Olympus OM-1 (and other Olympus and OM System cameras) is not limited to producing mushy images of running water and long tail-light tracks of moving vehicles in night-time photography. It can also be used to substantially reduce noise in low-light images of static subjects without the associated loss of detail produced by de-noising performed by the camera firmware.

The usefulness of Live ND in the OM-1 is somewhat reduced by the ISO 800 limit built into this feature. OM System probably has technical reasons for this choice.

I have nothing negative to say about this feature. It does substantially lengthen the total duration of the exposure, but this is just what the feature is expected to do.