Topaz Labs DeNoise 4.1
A New Photoshop Plug-in for Noise Mitigation
By Mark Segal
My first reaction when they announced it was kind of ho-hum, yet another up-date of a noise removal program – we already have at least three very decent ones (Neat Image, Noise Ninja and Noiseware), not to speak of the improved tools in Lightroom, Camera Raw and other image editing applications. But Topaz Labs claims it’s the best on the market and the download of a fully-functional 30-day trial is free, so why not give it whirl? I did and I was impressed. It’s easy to install (a plug-in in the Photoshop Filter menu, but also works with Lightroom, Photoshop Elements and several other applications), easy to use and the results (noise reduction with detail retention) are the best I’ve achieved from any noise mitigating application.
My interest in this is two-fold: minor - reducing noise in digital images; major – reducing grain and scanner noise from film scans. Why? Digital noise reduction is looking after itself these days with the latest crop of DSLRs – the images are more often than not so clean that it’s only under stress conditions (really high ISO, low light, underexposure) that an industrial-strength noise remover is needed. In fact, to test the application and prepare this review I took steps I’d never otherwise deploy in order to generate noisy images. Removing film grain while retaining image detail, however, remains a challenge for all those people scanning their archives and the small minority still shooting film. Applications designed for removing digital noise aren’t designed with scanner noise and film grain in mind, but if they can handle that too, it’s icing on the cake. So I tested both digital image files and colour negative film scans and report on those results here.
This review is not a step-by-step “how to” for using the plug-in. That’s all contained in the manual accompanying the download. It’s easy to follow and the GUI is very user-friendly – it has only a small number of options to tweak and in fact can be self-taught in little time. Rather my purpose is to show what it does, with some comment on the settings as we move along.
To create the noisy digital images, I went for my point-and-shoot – a Panasonic Lumix LX-1 (the first in a succession of this camera line), set the ISO to 400 (the maximum), selected an under-lit subject and under-exposed the image on purpose, then in Photoshop cranked-up the exposure to fully reveal all the noise. Wow! Did that ever work well! (Figures 1 and 2). Figure 2 shows a portion of the same image scaled to 100% screen magnification in the Topaz GUI. Notice the characteristic green/magenta speckles all over the wall, which is off-white in the real world, and how the titles of the books in the book case are partially obscured by noise.
Next step was to see how well Topaz DeNoise could improve this image.
Figure 1 – Noisy-1
Figure 2. Noisy-1, Detail – Uncorrected Preview
Topaz DeNoise has six sliders for removing noise, two for improving detail and edge sharpness, and one for adding grain. It also allows us to look at where the noise is, by providing RGB, luminosity, red channel and blue channel views. The first slider is for “Overall Strength” which controls everything else. The other sliders are for Highlight, Shadow, Red Channel and Blue Channel noise. The sixth is for cleaning-up clumps of colour noise. Once Overall Strength is set, the others work incrementally off of that setting. “Recover Detail” is for recovering image detail and texture, while “Reduce Blur” restores edge sharpness.
The manual warns repeatedly to proceed gently because the tools are powerful; I found indeed that the warnings are well-taken. One can easily overdo an effect. The idea is to get the balance just right between noise reduction and detail and edge retention. Detail Recovery can counteract Noise Reduction, but generally to a much lesser extent – that is, detail and edge recovery are usually far more effective than their counter-impact on noise reduction. The program comes with a set of RAW and JPG presets for those who like starting the adjustment process in that way. I don’t. I start from scratch by clicking the “Reset” button and then build the effect I want. I do all the noise reduction I think satisfactory, then move on to the Detail Recovery sliders and shift them until I think the balance is good. It is sometimes necessary to iterate between noise reduction and detail recovery to achieve a satisfying balance, but the process is fast and easy.
Unfortunately, there isn’t a split pane view to see “before” and “after” together, but clicking on the large image window flips one back and forth. Each time a slider is moved, there is a pause for several seconds while the program recalculates and displays the amended effect. One can undo, redo, make snapshots of settings, and create user-configured presets.
Because of how very noisy I made this image, it needed correspondingly aggressive treatment – intentional to test the strength and quality of the application.
Figure 3, Noisy-1 Corrected
Zooming in at 100% magnification, Figure 4 (left) shows an extract of Figure 3 as it appears in the Topaz interface.
Figure 4 – Noisy-1 Corrected, Detail (left) Figure 4-a Noiseware Comparator (right)
For comparison, Figure 4-a (right) shows the best result I was able to achieve using Noiseware on the same image1. Noise reduction and detail retention are not as good.
Apart from the extent of the overall image quality improvement in the Topaz result, what struck me most was that the texture of the lampshade survived very aggressive noise reduction thanks to the detail and edge recovery functions. Hence I decided to explore this capability in more detail by making another under-exposure at ISO 400 featuring the lamp shade, with the light off in order to maximize noise, as shown in Figure 5. Figure 6 shows the result of the clean-up, using the same approach as for Noisy-1.
Notice how the texture of the lampshade is reasonably well-preserved, while the noise has been almost extinguished. I was unable to duplicate this quality of combined noise mitigation and detail retention with any other noise mitigation application I currently have on my system.
An important additional consideration for the image processing workflow is that noise mitigation should be implemented before sharpening for prints or web output, as one wouldn’t want to sharpen noise before trying to mitigate it. As well, the noise reduction should be complete enough for the image to withstand subsequent sharpening.
1: I was not able to compare with Noise Ninja or Neat Image as I replaced those applications with Noiseware, but I consider all three to be fairly close contenders in terms of their capabilities.
Figure 5. Noisy-2; Figure 6. Noisy-2 Corrected
For example, Figure 7 (left side) shows the screen appearance of Figure 6 (“DeNoised”) once treated with PK Capture and Output sharpening. Figure 8 (right side) shows what the image would have looked like before using Topaz DeNoise but after the same sharpening. Notice in the left image the cleanliness of the background (it’s a wall) and the detail of the texture retained in the lampshade.
Figures 7 & 8, Noisy-2 Sharpened After & Before Noise Reduction
Turning to the use of Topaz DeNoise on film scans, scanning 35mm film with my Nikon Super Coolscan 5000ED using SilverFast Ai6 Studio provides the best quality film scans I can make. However, I’m not impressed with the results from any grain reduction tools which come bundled with scanning applications which I’ve tried. There is too much trade-off between noise reduction and detail retention, and not enough process control. This is all the more reason – at least for me – not to bake either grain reduction or sharpening results into a scan. One reduces the risk of having to rescan by implementing grain reduction and sharpening with specialized applications after scanning.
Figure 9 shows the whole photo I selected for this test, the black box indicating the area on which I concentrated for evaluating results. The scene is at Wat Arun in Bangkok (2004) shot with ASA 200 colour negative film; I selected it because of the combination of open sky and fine architectural detail which could combine to thwart creating relatively grainless skies while retaining essential subject detail. I scanned it at 363 PPI, 8.5 * 14 inch dimensions, using SilverFast AI 6 Studio with Negafix. The only image adjustment I made in SilverFast was to select the appropriate Negafix film profile. The resulting white balance, colour balance and exposure were good.
Figure 9. Wat Arun – Colour Negative Scan (De-grained)
The area in the black box (Figure 9) is shown in Figure 10, and the de-noised version in Figure 11 immediately below it. These are screen-grabs at 100% magnification.
I used the same workflow procedure in Topaz DeNoise as described above for the digital images, but the settings were less aggressive, based on need. Figure 11 shows a very slight residual amount of grain apparent at 100% magnification. I left it there despite the large amount of headroom which remained for more aggressive grain reduction, because my print size would be less than indicated by this extent of screen magnification, and it is good practice not to reduce grain any more than necessary regardless of what the software can do with detail retention, because detail recovery, good as it is, is not 100%. Also I found that grain reduction can be pushed too far, such that while retaining good edge detail, the image can lose just enough texture to have an undesirable overall “plastic” appearance.
The important items on which to focus in Figure 11 are the sky, the fine detail on the building façade and the definition of the red roof tiles. The improvement to the sky is very substantial and the detail retention on the building is very good.
I have also used Noiseware for reducing scanner noise and film grain in film scans. Figure 11-a shows the comparative result of using Noiseware rather than Topaz DeNoise on the same image as shown in Figures 10 and 11. Grain reduction and detail retention are less than achieved with Topaz DeNoise in Figure 11.
Figure 11-a Wat Arun Noiseware Comparator
I then decided to take it one step further and “max-out” the Topaz settings just to see how much soother I could make the sky and how much detail I could retain at that smoothness. The result of those operations is shown in Figure 12.
In Figure 12, the sky is smoother, and the detail looks slightly “smoothed”; the edge detail retention on the building is fine; the roof tile detail looks cleaner but perhaps a bit “plastic”. Whether one prefers the result in Figure 11 or Figure 12 is largely a matter of taste, the important point here being that the application can achieve either. This can be better appreciated seeing the original photos on a display, rather than as small JPEGs in a publication. Again, the results I’m achieving with Topaz DeNoise 4.1 exceed the best I achieved with previous noise-reducing applications I’ve used. Not that they were poor – this is just better.
Figure 12. Very Aggressive Grain Reduction
The real-world test of all this is to then sharpen the images and make prints. I printed both the less aggressive and more aggressive versions at 11*17 inch image size, 300 PPI on Ilford Gold Fibre Silk in my Epson 3800 printer. The results in print replicate what I see on the display: smooth skies and abundant detail. I’m also noticing a slight haloing along some edges, but not all, between building and sky. This would indicate either an image peculiarity or the need for more fine-tuning.
Finally, a comparison of Topaz DeNoise with the improved noise reduction tools in Lightroom Beta 3 is appropriate, as the latter will become a commercial release this year. Working with the same 3 images (Noisy 1 and Noisy 2 digital and Wat Arun film scan) in Lightroom Beta 3, I optimized noise reduction, detail retention and sharpening as best I could and the results are shown in Figures 13, 14 and 15, the left side being the corrected version and the right side the starting point.
Figure 14. Noisy-2 in Lightroom
Figure 15. Wat Arun Film Scan in Lightroom
One may compare the left-hand images of Figures 13 and 14 (digital) and 15 (film) with their counterparts in Figures 4 and 7 (digital) and 11 (film). My observations are (a) Lightroom Beta 3 noise reduction and detail retention for digital images is impressive compared with what was possible in Lightroom 2.7, and at more normal magnifications would produce very acceptable results; (b) Topaz DeNoise is marginally better, producing a somewhat smoother result with slightly better detail; (c) Lightroom noise reduction seems not particularly effective for film scans, if the results I achieved were typical, whereas Topaz DeNoise performed well on the same film scan.
To conclude, I think this is a very effective piece of software. I recommend readers to download a demo and try it out – it’s full-featured and free for thirty days, after which purchase is $79.99 (US).
Mark D Segal
Mark is a periodic contributor of essays and ideas to this website. He has been making photographs for five decades, got into digital imaging with Photoshop 6 and a film scanner in 1999, was fully digital and a "raw image" convert since 2004 and has produced thousands of inkjet prints using successive generations of Photoshop, Camera Raw and Epson professional printers.
A sample of Mark's work can be seen at www.markdsegal.com, where he also has information about custom training he provides on Photoshop, Photoshop Elements, Lightroom and SilverFast (scanning software)."