Turn Down that Noise

Comparing Three Noise-Reduction Programs– Which is Right for You?

By Ctein Back to

Ctein1so9

regularly use several general purpose noise-reduction programs to reduce the noise in my digital camera photographs, my film scans, and in the print scans I do as part of my restoration business. Noise reduction in all its varied forms is a staple of my work. I ’ve tested many and settled on three: Neat Image Pro+, Noise Ninja Pro, and NoiseWare Professional. (I did not test or investigate noise-reduction programs that use camera- specific profiles, of which there are many.) All are available as Photoshop plug-ins and stand-alone apps.

I use all three often enough to say they all have their merits. They all also have substantial differences, and you may very well find that one or more of them suit you not at all, or that one is so perfect for your needs that you don’t need to bother with the rest. It’s one of those cases where your mileage almost surely will differ from mine.

Customizability matters as much or more to me than the program’s default behavior and just what noise-reduction algorithms it uses. I’m always running into situations that require me to customize the noise-reduction settings to get the very best results, so how well a program lets me do that is extremely important to me.

Overly aggressive noise reduction tends to make everything look like it’s made of vinyl—smooth, shiny, sharp edged, and completely lacking in surface texture. The plastic look does not appeal to me at all. In almost every case, I didn’t care for the results I got using the software’s default settings. Most of the time, it was because the plug- in worked too well! It so thoroughly obliterated noise that it wiped out too much f ine detail and produced that “vinyl” look that I want to avoid. Hence, simple raw noise- reduction horsepower is a poor measure of quality; I care about how much noise I can eliminate without overly compromising subtle photographic detail.

I tested my three programs on three different cases: a low-noise, low-ISO digital photograph, an extremely noisy high-ISO digital photograph, and low-to-medium-noisy color negative f ilm scans. Before I discuss their performance, here are summary descriptions of the three programs, in alphabetical order:

Neat Image Pro+

Neat Image (www.neatimage.com, Figure 1) was surprisingly easy to use, considering how many control settings there are to play with. I barely had to glance at the quick-start instructions before diving in. That’s not an argument against reading the 60-page manual (you’ll get a lot more out of the program if you do); rather, it’s a compliment on how well Neat Image is designed. Controls have pop-up windows that give explanations of how to use them when you mouse over them.

To generate its noise profile, Neat Image analyzes a small portion of the image that is free from true subject detail to determine the grain and noise characteristics. If you don’t like its automatic selection you can move the selection box to a more appropriate location. Neat Image only analyzes one area in detail, but the Fine Tune button refines that prof ile based on the entire image.

When you preview the profile’s effects, you can alter the filter settings with separate sets of controls for low-, medium-, and high-spatial-frequencies. I frequently find it invaluable to be able to control how much noise reduction and post-reduction sharpening gets applied to different spatial frequencies. You can save the noise profile and settings if you want to use them on a series of similar photographs or experiment with a bunch of different settings to see which ones give you the most attractive final result.

(www.imagenomic.com/nwpg.aspx) is the huge amount of control it gives me over how noise reduction gets applied. You will need to read the manual on this one to use the program to its best advantage. It offers many, many controls for controlling noise reduction. Its best feature is “parameter bracketing,” which lets you select any one of the controls and automatically generate up to seven preview tabs with different values for that parameter. Figure 2 shows tabs that bracketed the sharpening strength.

This eliminates most of my trial-and-error fiddling. Comparing bracketed previews allows me to quickly decide what the best setting is for a parameter. Once I settle that parameter, I can pick a different one to bracket around and generate a new set of previews that incorporates the control settings I’ve established so far. In addition, there seem to be an unlimited number of undos, should I start fiddling with settings more casually.

Of all the noise-reduction tools I’ve played with, I feel this one gives me the most direct and flexible influence over the results. Because I have superior control over its behavior, I often get better results with Noiseware than any other noise-reduction program I’ve tried.

Noise Ninja Pro

Noise Ninja (www.picturecode.com/index.htm) is something of a standard in the field, especially amongst digital photographers, because it’s very good at creating noise prof iles for equipment like digital cameras and scanners. I use it because it often does a good job of noise reduction straight out of the box. The help manual for the program includes a 30-second guide to using it, which is actually sufficient in many cases. The five-minute guide will probably tell you more than you need to know, and hardly any of you will ever read the whole manual.

Noise Ninja (Figure 3) automatically multi-samples the photograph, usually with excellent judgment, but I recommend that you read the sections in the manual on how to manually profile a photograph and use the Brush tool. They’re not difficult to use, but you’ll want to use them most effectively.

Noise Ninja would be my favorite program if it gave me more control over how the noise reduction gets applied. For instance, it lacks controls to let me readily fine-tune noise reduction as a function of tone or spatial frequency.

The tests

The low-noise digital photograph is from a Fuji Finepix S100fs 11MP camera at ISO 100. I chose the scene to push the full 11-stop exposure range of this camera; it also turned out to be an excellent test of the trade-off between fine detail and noise. Not surprisingly, the degree and quality of noise varied hugely with the tonal level in the scene, from nearly none in the highlights to moderate in the deepest shadows (Figure 4).

Figures 4b and c illustrate the problem with too-aggressive default settings. The noise is gone, but so is too much of the texture in the fabric on the back of the chair; when the contrast drops below a certain level, it simply disappears in an unnatural- looking way. Figures 5 and 6 show the best results I got with Neat Image, Noiseware, and Noise Ninja in the shadows and the highlights. In this single test case, Noise Ninja performed best with its default settings. The other two programs performed better with custom settings, shown in Figure 7.

Overall, all three did famously, but the default Noise Ninja retained nicer shadow detail than the other two programs. On the other hand, it did a poorer job holding detail in the light midtones and highlights, where Noiseware was superior.

My high-noise test photo came from a Fuji Finepix S6000fd 6 MP camera at EI 6400 under very unfavorable lighting for color (Figure 8). This may be worse than any photo you’ll ever have to deal with, which is why I chose it for my tests. In this situation, none of the plug-ins produced acceptable results at their default settings. Neat Image and Noiseware suppressed almost all the fine detail, leaving a foggy-looking image; oppositely, Noise Ninja had a great deal of trouble automatically detecting the noise and eliminating it.

Custom settings (Figure 9) did a much better job. None of the plug-ins could entirely suppress the noise without sacrif icing excessive amounts of detail, but they all took it down to a bearable level. Noiseware arguably did the best job of suppressing grain and retaining real texture, as a result of massive customization of its settings (Figure 10). Noise Ninja did a much better job once I manually selected the regions for it to profile, but it still did the poorest job of the three. Not visible in Figure 9 is how the plug-ins handled the very-low-frequency chroma noise in the broad background areas: Noiseware eliminated it almost entirely, while Noise Ninja dealt with it poorly.

Finally there’s the matter of film scans, a subject of considerable importance to me as I have a substantial body of work on film. I tried all three plug-ins on several different photographs and didn’t get anything like consistent results. Even with the best custom-refining of settings, each of the three plug-ins proved markedly superior in some cases and substantially inferior in others.

Neat Image, in a few cases, produced better results with its default settings than the other two plug-ins did with customized prof iles. Usually its results look pretty good “out of the box.” Conversely, there were some photographs, like Figure 11, where it did the least satisfactory job of all three plug-ins.

Noise Ninja was easily confused by complicated film images; much as with the high-ISO digital photograph, it had trouble automatically finding the best areas of the photograph to build a profile from. It did much, much better if I selected the area manually. Then it often produced excellent results with less fussing than the other two programs. On the other hand, some scans gave it problems no matter what; it never did a satisfactory job of minimizing the highlight noise in Figure 11 unless I turned the strength high enough to seriously degrade fine detail.

The great adaptability of Noiseware meant that it was less likely to leave some noise unreduced or completely wipe out fine detail. With the other two plug-ins, there would frequently be some tonal values or some region of the color space where they didn’t function very well. Conversely, Noiseware rarely produced overall better results. There were always exceptions, however; Figures 11 and 12 represent a case where I feel that Noiseware did the best job of holding the visual qualities of the original while reducing the amount of grain.

Well, I doubt that you expected me to come to some sort of definitive conclusion about the superior program. After all, I said at the beginning that I used all three of these regularly. What I think sets them apart is the very different sorts of controls they give you over noise reduction. Collectively, there’s just about no noise problem I can’t tackle.


About the Author

Ctein
Ctein
Ctein is a technical writer and expert printmaker. He is also the author of Digital Restoration and Post Exposure—Advanced Techniques for the Photographic Printer.