Noise Reduction in Photoshop
As we learned in Chapter 2, “Why Do We Sharpen?”, noise or film grain is a fact of life for photographers. If you use a higher ISO or underexpose your image, the reduction of photons collected by the sensor or film results in more obvious perceptible noise. Shooting with a higher ISO often requires noise-reduction strategies in order to optimize an image for reproduction. Sometimes simply downsampling the image will be a sufficient noise-reduction tool. Other times you’ll need to deploy more aggressive reduction tools.
One thing to keep in mind, however, is that these are noise-reduction tools, not noise elimination tools. A certain degree of noise is expected or even desirable. If you totally eliminate noise or film grain, not only will the image look synthetic, but it will also have lost a large amount of actual image detail. That is the ultimate challenge of noise reduction: reducing the noise while preserving real image detail.
Just as it’s easy to overdo sharpening, it’s also easy to overdo noise and grain removal, leading to two undesirable conditions:
- Overly aggressive noise removal produces soft, unnatural-looking images that scream “digital” and, in extreme cases, appear to have been run through the Median filter.
- Many noise-reduction tools, including Photoshop’s Reduce Noise filter, have a tendency to produce images that aren’t as easily sharpened, because as soon as you apply sharpening, artifacts from the noise removal pop into view.
In many cases, the best way to handle noise is to simply protect it from being sharpened. But in extreme cases—for example, high-ISO or heavily underexposed digital captures, or scans from high-speed film, especially color negative—you may have to reduce the noise before you start sharpening.
The Reduce Noise Filter
Photoshop’s Reduce Noise filter can do a creditable job, though as noted previously, frankly it’s not as good as most third-party plug-ins. The danger with Reduce Noise is that you wind up creating an image that can’t be sharpened. Figure 4-35 shows an example.
Figure 4-35 Reduce Noise dangers
Reduce Noise Advanced Options
As with Smart Sharpen, clicking the Advanced button changes the interface and brings up a Per Channel option to apply differing amounts of noise reduction based on the channel. This is useful for film scans, as the Blue channel is often noisier than the other channels. As with Smart Sharpen, however, the settings you apply when in the Advanced mode continue to be in play when you select the Basic button. For this reason, we suggest always using the Advanced mode. Figure 4-36 shows the Advanced options.
Figure 4-36 Reduce Noise Advanced
When selecting the Advanced options, the Per Channel settings override the Overall settings for Strength and Preserve Details. The only Overall settings that will be applied will be the Reduce Color Noise, Sharpen Details, and the option to Remove JPEG Artifact. If you process raw files from Camera Raw or Lightroom, the optimal place for doing color noise reduction would be in that processing pipeline. Reapplying color noise reduction in Reduce Noise is not suggested, as it will potentially blur the color data and leads to color bleeding across color boundaries.
Saving Reduce Noise Presets
Like Smart Sharpen, Reduce Noise allows you to save commonly used settings as a preset to be applied more efficiently. Figure 4-37 shows the Save and Trash icons in the Reduce Noise dialog box, as well as the New Filter Settings dialog box and the Settings drop-down menu. To save settings, click the Save New Filter button and name the saved setting, which will then appear in the drop-down menu.
Figure 4-37 Reduce Noise saved settings
Strength vs. Preserve Details
As with all noise-reduction tools, the primary challenge is to reduce the noise while preserving the edge detail in the image. We strongly suggest applying Reduce Noise on only a copy of a pixel layer so that you can adjust the opacity and create your own surface mask to more accurately attack the noise and preserve details. The technique for creating the layer and mask is essentially the same as that for a sharpening layer with an edge mask, only the edge mask for noise reduction isn’t inverted, so the edges show as black and hence are protected from the noise reduction.
The Surface Blur Filter
Tucked into the Blur filter group is a little known and rather underused blur filter that’s useful for reducing noise. The filter has a Radius setting and a levels threshold. Combined, the settings allow the blur to be applied on the non-edge areas of an image—hence the name Surface Blur. It should be noted that this is a rather un-subtle tool. It’s very strong; used too strongly, it can quickly turn your photographic image into an impressionistic mess. However, used with a degree of finesse and with the addition of a surface mask, Surface Blur can reduce some of the most stubborn noise, particularly noise that is amplified by strong boosts in post-processing exposure adjustments.
Figure 4-38 shows the Surface Blur dialog box as well as the surface mask we’ve created for use with Surface Blur and Despeckle.
Figure 4-38 Surface Blur and surface mask
The key to using Surface Blur successfully is to localize the application both by using a surface mask as well as using reduced opacity. We’ve often found that applying Surface Blur and applying the result at 50% opacity allows a reduction in the perception of strong noise. The noise is still there because the opacity is reduced. But the gentle use lessens the strength of the naturally present noise.
Noise Reduction with Despeckle
The Despeckle filter is useful for dealing with mild-to-moderate luminance noise (as opposed to color noise). The secret is to apply Despeckle differentially to the three color channels—the red channel usually has the least noise, and the blue channel usually has the most.
Figure 4-39, below, shows the result of this technique, which is a simple filter menu command (Filter > Noise > Despeckle), in action along with additional noise-reduction tools. When overdone, this technique can produce color shifts, but it provides an easy way of knocking back some of the worst of the noise. As with all sharpening and noise-reduction tools, it’s best applied to a copy of the image.
Figure 4-39 Comparing noise reduction at 400% zoom
Comparing Noise Reduction Tools
In Photoshop, there’s often not a single “perfect tool” you can call on to solve your imaging needs, and this is certainly true with noise reduction. Many times you’ll need to combine multiple tools to achieve optimal results. In this section we’ll compare the preceding noise-reduction tools as well as a couple extras: a third-party tool (Noiseware™ from www.imagenomic.com) and Photoshop’s own Camera Raw. There will be two sets of comparison images, one set for pixel-peeping at a screen zoom of 400% and a contact-sized print set to 300 ppi and sharpened for output. The original, an ISO 1600 digital capture from a Digital Rebel XT, is a 6 MP capture and would reproduce at 11” x 7” at 300 ppi.
Comparing at 400% Zoom
Figure 4-39 shows an array of six images of a small portion of the original capture. At 400%, the image on screen would be effectively 12 times the actual size (at 100% of a 100-ppi display, the image would be three times the original, so at 400% it would be an additional four times bigger). So while what we are seeing is accurate, it is not at all representational of final output reality. In fact, viewing these images in the book is even further removed from reality, because the results have been put though the halftone screening process to produce 150-lpi halftone dot.
There are some interesting conclusions to be drawn from this comparison. First, looking at the noise results of image processing at a 400% zoom isn’t particularly enlightening. Yes, you can see what the pixels look like, but you can’t really see what they mean in terms of signal-to-noise. Camera Raw, which is often criticized for inferior noise reduction, fares pretty well compared to Reduce Noise. Surface Blur with a surface mask does a good job (better than Despeckle) of substantially minimizing noise without too much obliteration of image detail. The third-party plug-in at default appears best, which isn’t surprising—as with all things in life, you tend to get what you pay for.
Comparing at Contact-Print Size
In the grand scheme of things, what really matters about noise reduction is what do the results look like in the real world? Going back to the concept of a contact print, introduced in Chapter 2, the following images are crops of the full frame at 300 ppi, then output sharpened for this book’s 150-lpi halftone screen. Let’s see if these contact prints confirm or conflict with your impression of the noise-reduction tools’ results when viewed at the previous 400% zoom.
Figure 4-40 shows the contact-print size of each of the noise-reduction examples.
Figure 4-40 Comparing noise reduction at contact-print size
If you’re not quite sure what to think about this comparison, let us take this opportunity to tell you what we think. We think that by the time the image has been processed, cropped, and output sharpened for reproduction, the differences between these results are very subtle indeed. We would argue that, in many respects, the image processed in Camera Raw has perhaps the best combination of noise reduction without artificial, detail-losing smoothing. It looks more photographic.
We also think that Reduce Noise is a fine and dandy tool if used well, but if you have a few extra shekels, investing in a third-party noise-reduction tool is well worth the money. While we use Noiseware, that’s not to say that any of the others are not equally useful when compared to Photoshop’s toolset.
In extreme noise situations, many third-party plug-ins do a better job than can be achieved using Photoshop’s tools. ABSoft’s Neat Image, Visual Infinity’s Grain Surgery, PictureCode’s Noise Ninja, and Imagenomic’s Noiseware Professional all provide industrial-strength noise reduction with a great deal of control over the process. (If we’ve failed to mention your personal favorite, it’s simply because the aforementioned plug-ins are the ones with which we’re most familiar.)
Here are a few general guidelines for using third-party noise-reduction solutions:
- Always do noise reduction before sharpening unless the sharpening and noise reduction are integrated in the same processing pipeline. If you sharpen first separately, you’ll almost certainly make the noise worse. The noise-reduction tool will have to work harder, and will probably wipe out the sharpening you did anyway.
- If you perform noise reduction on a layer, you can reduce the noise slightly more than you actually want, then fine-tune the noise reduction by tweaking the layer opacity.
- Don’t overdo the noise reduction. A certain amount of noise is usually preferable to an image that looks like it’s been blurred.
In practice, we find we need to resort to third-party tools only with high-ISO (1600 or greater) or severely underexposed (more than 1 f-stop) digital captures, or with scans from color negative. In virtually all other situations, the techniques presented here work well.
Ironically, it was during the preparation of these images for the book that a key aspect of noise reduction and image sharpening sort of fell into place—the ideal place to accomplish the noise reduction is in conjunction with and in the same process as the capture sharpening. This isn’t such an earth-shattering revelation, since Bruce had advocated something similar in the first edition of this book. But the principal has been driven even further home now: If at all possible, do your capture sharpening and at least the primary noise reduction in your raw processor, particularly if that raw processor happens to be Camera Raw or Lightroom.