Framed and Exposed: Quiet Those Noisy Images

Rating: 
Body: 

Noise -- those ugly speckled patterns that can sometimes look like film grain and sometime just look ugly -- is a concern for all digital photographers, but as digital cameras have improved, noise has become less of an issue. Nevertheless, all photographers keep an eye on noise in their images, and most have some plan of attack at the ready for removing noise when it crops up.

In this overview, we're going to take a look at 3 packages side-by-side, evaluating them for their ability to handle images produced by a digital SLR. Because digital SLRs have much better noise response than lower-end digital cameras, they typically require much less noise reduction. Unless you spend lots of time regularly shooting at high ISO with your SLR, you probably find that you only occasionally need to employ a noise reduction technique, and so want a simple, quick-and-dirty, fix. All three of these applications are well suited to users who only occasionally need to process a noisy image, as well as users who want to make noise reduction part of their everyday workflow. In addition, all three work on either Macintosh or Windows.

Noisy nature
Noise comes in two flavors: luminance noise and chrominance noise. While all noise can be annoying, luminance noise is the preferable of the two. Luminance noise presents itself as speckly changes in lightness and darkness and often looks very similar to film grain. Chrominance noise is the more typical "digital-looking" noise and appears as colored blotches, usually magenta or green. These discolorings almost look like small stains scattered all over your image.

Digital SLRs typically suffer more from luminance noise than chrominance noise, though in some cases images can have both. In general, even when chrominance noise does occur in an image from a D-SLR, it's much less pronounced than in a lower-end digicam.

When reducing noise using any technique, your main concern is to eliminate noise without reducing detail in the image. In addition, images can loose sharpness as pixels along hard edges are eliminated or reduced, and sometimes colors will shift or the image will develop an overall cast as contrast between adjacent pixels is reduced. All of these products are capable of trashing image detail -- resulting in posterized colors and an overall "lower-res" look. Fortunately, they each provide varying degrees of control that allow you to balance noise reduction, detail loss, and color shift.

Because SLRs tend to yield less noise overall, and usually only suffer from luminance noise, noise reduction of SLR imagery is a somewhat easy task.

Noise Ninja
PictureCode Noise Ninja is a $69 package that comes as a stand-alone application or Photoshop plug-in, and offers the best balance of price and performance of the three packages covered here. Noise Ninja is a profile-based system, which means that it can be configured to deal specifically with the noise characteristics of your particular camera. Noise Ninja's profiling system is very well conceived and easy to use, and the program can generate a rough profile of your camera from the image that you're trying to de-noise.

20050222_fg1.jpg
Figure 1: Noise Ninja's single-dialog interface.

To use Noise Ninja, you open your target image, then select an existing profile, or tell the program to generate a profile on its own. The program can generate auto profiles very quickly -- the program only required 13 seconds to generate a profile for our our 2048 x 3072 pixel test image, when running on a 1.5 GHz G4.

Once the profile is locked in, you're ready to configure the noise reduction parameters. You'll want to work at 1:1 (100%) size when defining your noise reduction parameters. Noise Ninja provides a rectangular test area that you can move about your image to see the effects of your current setting. This mechanism provides for good before and after experimentation.

The noise controls themselves are fairly simple. Luminance, Color, and Sharpening sliders let you target luminance noise and color noise, while simultaneously performing an unsharp mask operation to try to improve lost image detail. An Advanced Controls option provides additional sliders for each parameter, allowing you to fine-tune your configuration even further. Finally, color-specific controls let you target specific instances of color noise blotchy-ness.

As you can see in the following figure, Noise Ninja does a very good job of smoothing out noise without compromising too much detail.

20050222_fg2.jpg
Figure 2: On the left, a 1:1 crop of our original image. On the right, the same crop from our Noise Ninja'd image.

These images were created using the program's Auto profile feature. We reprocessed the same image using the custom Canon 10D profile, available on PictureCode's web site, but saw little or no difference in result. As you can see, the program not only tackled the slight luminance noise, it also eliminated the subtle chromatic noise visible in the background shadow. Though the crop shown above shows an acceptable level of detail loss, other parts of the image didn't fare as well, as can be seen by this crop.

20050222_fg3.jpg
Figure 3: A different part of the same image filtered using the same parameters as the previous example.

As you can see, skin texture has been lost, resulting in the skin having a "plastic" quality. In addition, the skin tones have shifted a little more yellow from the loss of some of the scrubbed tones.

Fortunately, Noise Ninja provides an excellent solution to this problem in the form of a Noise Brush that allows you to apply selective noise reduction to specific areas. In general, you'll probably find that you want to perform less noise reduction in areas of high detail such as hair, foliage, and other "noisy" subjects. Any noise reduction program is going to have a difficult time differentiating between content and noise when dealing with such areas of the image.

Other niceties
Though Noise Ninja adds another application to your image processing workflow, it's speedy enough that having to drop into a separate app to perform noise reduction is not a terrible inconvenience. What's more, the program provides some extra goodies. It's multi-processor aware for better performance when using auto-profiles; internally, it's completely 16-bit, and provides support for 16-bit output; and it includes a batch processor for denoising entire folders of images using a specific de-noise parameter set.

What's more, Noise Ninja provides a simple method for profiling your specific camera. Shoot a picture of the included profile test chart and feed it to Noise Ninja, and the program will build a noise profile specific to your camera. Because noise response can vary between individual units, and because camera temperature and ISO have a profound impact on noise, you can build custom profiles tailored to your specific camera and shooting conditions -- ideal for users who shoot in particularly hot or cold environments.

If $69 seems a little steep, the $29 Home version provides the same interface and noise reduction mechanism, but lacks 16-bit support and batch processing.

Fred Miranda ISOx Pro
Constructed as a Photoshop Action, ISOx Pro, available from fredmiranda.com, provides a sophisticated level of noise reduction with the advantage that can be incorporated into existing Photoshop actions and procedures. Rather than relying on profiles, Miranda simply provides different versions of their software for specific cameras. At the time of this writing, custom ISOx plug-ins are available for the Canon EOS 10d, 1Ds, 1Ds Mark II, 1D, Digital Rebel, D60, D30, G-series, the Nikon D70, D100, D1x, Coolpix series. Finally, a "generic" ISOx Pro version is provided for users of other cameras. Each plug-in costs $15, except for the "generic" version, which goes for $20.

The downside to this approach is that if you upgrade your camera, or shoot with multiple, different cameras then you're going to have to buy several versions of the software. What's more, you can't build a profile for your specific unit. However, the differences between one version and another can be very subtle, so this may not be too much of a concern. Obviously, if you have several different cameras, and a limited budget (since you're out of money after buying all of those cameras) then you might be best served by the generic ISOx.

After opening an image in Photoshop, you invoke ISOx from the File>Automate menu. This brings up a single dialog that varies slightly depending on which version of the software you're using. For the purpose of this test, we ran the generic ISOx Pro version.

20050222_fg4.jpg
Figure 4: ISOx Pro provides a simple single-dialog interface within Photoshop.

ISOx's controls are very simple. A set of ten radio buttons provides ten levels of noise reduction. In addition, you can target the function to only attack chrominance noise, and perform sharpening on just the luminance channel to try to restore lost detail.

After clicking OK, the script will take off and begin processing your action. Though it's impossible to open the script to look inside, you can get a reasonable idea of what it's up to simply by watching the different processes that take place on the screen. ISOx seems to achieve most of its results by converting your image to LAB color, and then performing independent custom Smart Blurs on the luminance and color information. After this, your image is restored to RGB and some edge-protected functions of some kind are performed. In theory, these types of mode changes can introduce some data loss to your image through rounding errors during the color space conversions. However, it usually takes many many conversions before this data loss becomes visible, so these conversions should not have any impact on your final image quality.

Shown below are three crops showing original data, and two levels of ISOx noise reduction.

20050222_fg5a.jpg20050222_fg5b.jpg20050222_fg5c.jpg
Figure 5: Original image, Level 1 reduction, Level 2 reduction.

As you can see, ISOx did a slightly better job of preserving skin texture than did Noise Ninja, but it didn't reduce noise as much overall, and had a slightly more difficult time with the chromatic noise in the shadows.

Obviously, ISOx does not provide as fine a level of control as Noise Ninja. In addition, it's much slower to process an image. If you've got many images to de-noise, you might find Noise Ninja's faster processing to be a bigger workflow boon than Photoshop integration. However, if your noise reduction needs are simple, it's hard to beat the price.

nik Multimedia Dfine
At $100 for the software plus $40 for each camera profile, nik Multimedia's DFine Photoshop plug-in is far and away the most expensive noise reducer covered here. However, it also offers far more control than any noise reduction product we've seen. nik Multimedia is well-known for their sharpening software (nik Sharpener Pro) as well as several other Photoshop plug-ins. DFine packs the same excellent design as their other products, and provides a thorough, detailed manual that explains their approach to noise reduction.

If your camera allows it, nik recommends deactivating or minimizing your camera's internal noise reduction functions as much as possible. They also advise you to reduce your camera's internal sharpening functions as much as possible. Their reasoning is that their algorithms are more sophisticated than the relatively simple noise reduction and sharpening routines found in most cameras, so it's better to have an image that's as raw as possible to preserve the most detail information.

nik also believes that different types and levels of noise reduction need to be used for different colors within an image. Their camera settings contain complex noise-to-color profiles that, when combined with the user's input allow you better control of both luminance and chrominance noise reduction.

DFine's single dialog provides a configurable preview display that let you see before and after crops side-by-side or stacked vertically. Four buttons on the side let you display controls for Luminance Noise, Chrominance Noise, Contrast & Light, and Color Cast & Color Balance. Clicking on any one of these buttons brings up controls specific to that function.

20050222_fg6.jpg
Figure 6: DFine's interface provides good side-by-side previews and simple push button configuration

Once you select a parameter to edit (in the case of the dialog shown above, we're currently configuring luminance noise reduction) you can then select a method. Selecting Quick Fix gives you four levels of noise reduction to apply, with no control of customization. An array of Print Optimization choices let you opt for noise reduction routines that are optimized for preserving detail for printing. Finally, if you have a camera profile installed, you can select it to get a range of sliders like those shown above. Each slider represents a different color range, allowing you to specify different noise reduction amounts for different color ranges in your image. (Note that you don't have to have a camera profile to use DFine, but you will need it if you want to be able to target varying luminance reduction to specific color values.)

After configuring Luminance Noise reduction, you can move on to Chrominance Noise, which presents similar controls but with a different choice of methods. By choosing from four options, you can elect to perform a global reduction of chrominance noise, with a slider control to select the degree of reduction, or a Protected mode, which attempts to protect areas of color from the types of "flattening" we saw in earlier examples. Finally, the program offers a special JPEG Reduction mode, which attempts to remove JPEG artifacts.

20050222_fg7a.jpg20050222_fg7b.jpg
Figure 7:On the left is our original image, on the right is the DFine-processed result. (We were working with a demo copy, which is fully functional, but watermarks every image with a repeating "Demo" stamp.)

As you can see, DFine's extreme level of control allows you to target varying degrees of both chromatic and luminance reduction to specific areas of your image. While targeting the dark shadowy background to both luminance and chrominance reduction, we were able to protect the skin tones to prevent posterization (although some loss of tone still happened).

DFine takes longer to configure, and takes a good 45 seconds to a minute to process its final effect, even when using its "Quick Fix" modes. If you need to process a lot of images, DFine may not be the best choice. Similarly, if you only occasionally face noise reduction troubles, then the $100 price tag (or more if you want a camera profile to go with it) might be a little steep. If you regularly face very tough noise reduction tasks, though, DFine's thorough degree of control might make it the best option, if you can afford it.

In a future column, we'll be testing these same products with point-and-shoot cameras, which typically present much more taxing noise reduction problems.

Read more by Ben Long.

Liked This? Read These!

Nik Software Inc. today announced the latest upgrade to its award-winning noise reduction tool Dfine® 2.0. Designed to improve image quality, Dfine 2.0 now includes Nik Software's patented U Point... Read More
Nik Software Inc. today announced that its award-winning noise reduction software tool Dfine 2.0 is now available for sale from specialty camera retailers in the U.S. or direct from Nik Software (www... Read More
Dynamic range expansion by blending several images taken at different exposures is well known, adopted by photographers and implemented in a number of software tools. However, while the exposure... Read More
PictureCode released a plug-in version of Noise Ninja for Apple's Aperture photo editing and management software. The plug-in allows Aperture users to employ best-of-breed noise reduction without... Read More