Unless you've been living under a rock, you've probably heard about the DCS Pro 14n's Noise Problem. The camera has gotten a bad noise rap in part because of a misunderstanding of how to use Photo Desk's Noise Reduction function by some of the folks evaluating the 14n, and in part because the 14n is a noisy camera.
DCS Pro 14n, ISO 80, Adv. Noise Reduction: Radius 4, Strength 100% (Zoom | Full-Res)
Whether the quality and quantity of noise in a 14n photo will be acceptable to you will be dictated by what you shoot. A bride in a white dress against a white background, in a scene lit with many watt-seconds of light, is a potential 14n candidate, especially at ISO 80. Shooting the same bride's wedding ceremony in available light from the balcony of an ornate, dimly-lit church, on the other hand, will produce an unacceptably noisy result, or one that restricts your ability to enlarge the photo much beyond 5 x 7. Especially if the picture is taken at or near ISO 400.
As with any other digital SLR, blanket phrases such as "You can't shoot over ISO 125" don't apply. Also as with any other digital SLR, the visibility of noise is driven principally by the scene's tonal range and the ISO setting. The two wedding examples make this point: a high-key scene captured at ISO 80 will show relatively low noise (though it will still have some, possibly coupled with blotchy, posterized shadows). A low-key scene at the camera's current upper end of ISO 400 will be overrun with noise.
At ISO settings in between your mileage will vary, depending on how shadowy the environment is you're trying to photograph. Like all digital cameras, the proportion of dark tones in the frame drives the visibility of noise. Unlike any other digital SLR we've ever shot with, however, the visibility of 14n noise increases dramatically when the frame is particularly shadowy. The ISO 80 Golden Gate Bridge photo above, for example, is fairly noisy in the shadows of the bridge, even with the maximum possible noise reduction applied in Photo Desk.
Shooting the same San Francisco landmark after sunrise gives a somewhat less noisy result, in part because the large, flat areas in the photo are tailor-made for Kodak's noise processing, though there are still areas that show a noise blotchiness.
DCS Pro 14n, ISO 80, Adv. Noise Reduction: Radius 4, Strength 100% (Zoom | Full-Res)
EOS-1Ds, ISO 100 (Zoom | Full-Res)
Conversely, the snapshot below doesn't look that bad for ISO 400 from this camera, though the noise levels are still both unacceptable and more pronounced than perhaps any other current digital SLR. The shadow areas are rife with noise, but most of the picture isn't that shadowy.
DCS Pro 14n, ISO 400, Adv. Noise Reduction: Radius 1, Strength 0% (Zoom | Full-Res)
Note: The full-resolution version linked beneath the photo has been processed with the minimum possible noise reduction; to download the same photo with the maximum possible Photo Desk noise reduction applied instead, click here. To see three crops from the same photo processed with Medium, High and the maximum possible noise reduction, click here.
To understand why the 14n can be so noisy, it's important to look at one aspect of how image sensors work. It's an immutable trait of sensor design that, all other things being equal, larger pixel sensors have a higher signal-to-noise ratio than smaller pixel sensors. Put in photographic terms, smaller pixel sensors tend to produce noisier photos than ones captured with larger pixel sensors. Camera manufacturers do everything possible to work around this. By employing sophisticated noise reduction algorithms during image processing, as well as by designing or choosing sensors that use light-intensifying microlenses over each pixel, it's possible to generate an image that has much less visible noise than it otherwise would have at a given ISO.
But there's no changing the fact that as pixels get smaller, the proportions of noise and signal (where the signal is the scene you're capturing) change in favour of the noise. Canon had to tackle this in the design of the EOS-1Ds and its CMOS sensor's relatively small 8.9 x 8.9 micron pixels. So did Kodak, though their task was potentially greater: the 14n's pixels are more compact, at 8.0 x 8.0 microns. While Canon opted for microlenses over each pixel, Kodak did not. The net effect is Kodak faced a particularly big challenge in turning 14n sensor data into a clean, usable photograph, though by not including microlenses, they also reduced the likelihood of sensor-induced chromatic aberration. This is true in theory, and in practice, with the 14n.
In our discussions with Kodak, it's clear that they were counting on two things:
- By co-developing the sensor with FillFactory, Kodak would reap the signal-to-noise ratio benefits of Fillfactory's patented CMOS N-Well pixel design. This design increases the fill factor of each pixel, in effect allowing the sensor to gather more light (ie photons) per pixel than a traditional CMOS sensor. Fill factor is almost certainly a big deal to a company named FillFactory.
- That noise reduction processing would take the image the rest of the way there.
The 14n is a low ISO camera. If there is a blanket statement to be made about the 14n, it's that it will deliver its best results at bigger enlargement sizes only when the ISO is as close to 80 as possible. In fact, if you can accept the loss of a small amount of dynamic range, shooting this camera at ISO 50 or 64, then correcting the overexposure in Photo Desk (the 14n can't be set to shoot below ISO 80), will produce the cleanest files. That's how I shot both the 14n and 1Ds versions of the photo below. Each has had -0.3 software exposure compensation applied before conversion from RAW. Do the math and you'll see that makes the 14n frame approximately ISO 64, the EOS-1Ds frame approximately ISO 80.
This is not a moody available light camera at any ISO, and certainly not at the ISO 400 usually needed to shoot in moody available light. Add to that the fact that even at ISO 80 it's possible to pick up an unpleasant amount of noise in silhouettes and dark backgrounds, and what you have is a camera that will produce clean files only under a limited set of circumstances. The only real mitigating factor is the size of the print, because at 8 x 10 and smaller moderate amounts of 14n noise are muted considerably.
It should be pointed out that the EOS-1Ds doesn't produce a perfectly clean file either under the same circumstances, but by and large it will produce one that is more usable, and certainly more enlargeable. And, in available darkness it easily eclipses the 14n at ISO 400, which produces completely unusable results. The full-resolution versions of the photos below will give you an idea of what each camera does under fairly difficult conditions: a scene with lots of dark tones and an exposure of 1/2 second.
Note: In firmware version 4.1.2, 14n exposures up to 1/2 second are handled by one noise reduction method, 1/2 second to 2 second exposures are handled by another. In an upcoming version of firmware, Kodak intends to move this switchover point, which should improve 1/2 second exposures to some extent. To catch a glimpse of the future, download this 3/4 second exposure of the same scene. Though a bit brighter overall, and therefore inherently less noisy, the 3/4 second frame is measurably cleaner overall.
And finally, here's a look at the ISO 400 middle ground. The flash-lit areas of the photo aren't horrendously noisy, while the background and other shadow areas are. But in a 5 x 7 print the noise is far less noticeable than it is on screen at 100% magnification.
DCS Pro 14n, ISO 400
Adv. Noise Reduction: Radius 3, Strength 80% (Zoom | Full-Res)
14n noise is more than just noisy. Put another way, all noise isn't created equal. For instance, the EOS-1Ds at its upper ISO limit of 1250 can get noisy, but because the noise is mostly random and grain-like in appearance we don't find it particularly objectionable. At any ISO, 14n deep shadow noise has a structured, posterized look. This, more than any other noise characteristic, detracts from any 14n photo with significant shadow detail.
In the full-resolution versions of the ISO 80 photo below, examine the road entering the frame from the left, and other similar areas. The EOS-1Ds frame looks, and prints, with more natural and photographic gradation in the shadows. The 14n has an unnatural, plugged look to the shadows by comparison.
Getting to know Photo Desk Noise Reduction is essential. To keep noise at bay, it's critical to learn how Photo Desk's Noise Reduction function works, what the optimal settings are and to get a feel for when switching from Advanced to Advanced with Moire Reduction is going to be necessary.
Photo Desk Noise Reduction dialog
Kodak has taken a small step towards making the Noise Reduction defaults semi-automatic, by having the reduction applied be ISO dependent. In other words, even when one of the default settings of Low, Medium or High is selected, the noise processing applied on each setting varies with ISO. That doesn't eliminate the need to fine tune the settings for almost every photo, however, but it does help define the starting point.
The team photo test frame below, shot at ISO 100, is about the cleanest we've taken so far with the 14n, at least without applying software exposure compensation trickery. This is, in part, because the picture and Photo Desk's noise reduction have cooperated nicely, allowing for a fairly aggressive noise reduction setting without chewing up too much image detail.
DCS Pro 14n, ISO 100, Adv. w/Moire Noise Reduction: Radius 6, Strength 50% (Zoom | Full-Res)
Processing to reduce noise also reduces detail. We've already covered this in the Detail section, but it bears repeating: Photo Desk's Noise Reduction function tends to smear detail in low contrast areas. The higher the ISO, and the higher the Noise Reduction settings, the more pronounced the effect.
With previous models, such as the DCS 760, it was possible to handle noise reduction outside of Photo Desk, thereby avoiding the smearing effect. Noise reduction is always on during the conversion of 14 files, even when saving out ERI-JPEGs in the camera, so there is always a risk of detail being smeared. And because it's always on it's impossible to say if turning it off would introduce other image defects that noise reduction currently masks. But we suspect it would be nice to have the option.
Noise is muted by enlargement size. Also mentioned several times is the notion that 14n noise, while often painfully apparent when viewed at 100% magnification in Photoshop, isn't nearly as noticeable in prints of up to about 8 x 10 inches. Noise can be managed, in part, by restricting the print size of images that are low-key or have been shot at higher ISO settings. Of course, this might also be an argument for choosing a lower resolution but less noisy camera.
"Because we can, we will." Kodak's Steve Noble promises that the programmable, upgradable architecture of the 14n, both its firmware and Photo Desk, will allow for improvements in noise reduction filtering in the months ahead. Ongoing tweaks to current DCS models has been one of the hallmarks of Kodak dating back to the DCS 520, so there's reason to believe that the DCS group will implement changes to their noise processing algorithms for the 14n.
But it's not as if Kodak has done nothing through the camera's development cycle to counteract the inherent noisiness of the sensor, including the 14n's multi-second startup time while it builds a series of dark frame maps that it can use for noise reduction on the fly, its recalibration routine when changing ISO across analog gain points and the aggressive noise reduction already offered in Photo Desk. Somewhere along the way, Kodak engineers figured out this was going to be a noisy camera and took steps to beat back the noise during image processing.
While it seems likely, then, that Kodak can tune 14n noise reduction to better retain image detail, as well as implement minor tweaks to its overall effectiveness, the big picture will almost certainly remain the same: the 14n and its current sensor will continue to capture an overly noisy photo with odd, posterized shadows that will be best or only suited to well-lit, low-ISO shooting.