Topaz Labs DeNoise 4.1 Review: A New Generation of Noise Reduction Software
Product Category: Noise Reduction Software
Manufacturer: Topaz Labs
Requirements: Macs: Intel-based Macs with OS 10.4, 10.5 or 10.6 (Topaz is NOT compatible with PowerPC processors - like G4 or G5). 1 GB RAM minimum. Adobe Photoshop CS3-CS5, Adobe Photoshop Elements 6- 8. Apple Aperture 2 and 3, Lightroom 2 and 3, and iPhoto now supported via Topaz Fusion Express on OS 10.5 or higher.
Windows: Windows XP, Windows Vista, or Windows 7 (32-bit and 64-bit). 1 GB RAM minimum. Adobe Photoshop 7-CS5, Adobe Photoshop Elements 1-8, or another editor that supports Photoshop plug-ins. These include: Irfanview, PaintShop Pro, etc.
- Good quality noise reduction; on par or exceeds previous generation software;
- Easy-to-use interface.
- Slow; really slow;
- Settings can produce somewhat inconsistent results.
Summary: Noise reduction software is not entirely new; back in 2003 I first heard about Neat Image, and for a period it was Neat Image vs. Noise Ninja. Now, a new generation of products has arrived, and Topaz Lab's DeNoise 4.1 leads the pack. How does it do against an old stalwart like Noise Ninja? Read on!
As any enthusistic photographer knows, noise can be a big problem, especially when one starts increasing the sensitivity of the imaging sensor. Noise destroys detail and information, making photos look like a mess of dots and colour patches.
Well, the good thing about digital is that there is usually software to solve this problem. Cameras have always had some noise reduction built-in, but the processing power inside a camera just cannot match up with that on a dedicated stand-alone computer. As such, there is a range of software dedicated to removing noise from images.
I first heard about them from a professional photographer friend of mine; this was back in 2003 or 2004, and Neat Image was the program's name. Back then, cameras were quite a bit noisier, and the likes of Neat Image were quite welcome. I purchased a copy of Noise Ninja - Neat Image's main rival.
Times change, and Noise Ninja seems to have seen few updates (the last major version, 2.0, was released in 2006; the last update was 16 months ago), leading to some new challengers. The most prominent of these has to be Topaz Labs' Topaz DeNoise, which currently is in its fourth version. I decided to take it for a spin, and see how it stacks up to the old workhorse that is Noise Ninja.
Note about noise
Before I continue, here is a crash course about the different types of noise:
Luminance Noise: These are the little coarse specks that break up the image. Being too aggressive with removing this kind of noise can lead to plastic-looking images devoid of detail.
Chrominance Noise: "Chroma" noise shows up as those colour splotches all over the image. Generally most noise reduction software does a pretty good job at this - even Nikon does a great job of this in-camera for the past two generations!
Readout Noise: These usually manifest as "bands" in areas of near-black levels. This happens in high-speed cameras using sensors that have multiple readouts to move data quickly off the sensor.
Topaz DeNoise is basically a plug-in for PhotoShop. If your image editor supports the PhotoShop Plug-in format, you should be able to use it. Upon opening the plug-in from the Filters menu, this is what you get:
Figure 1: The Topaz DeNoise Interface (click for larger image).
On the left are the setting presets, along with the options for saving, exporting and importing them. In the middle is the preview pane, and on the right are the controls for setting the amount of noise reduction and detail management (noise reduction invariably will reduce some of the information present in the image).
The controls took some getting used to for me. As a long time user of Noise Ninja, I'm more familiar with its concept of giving luminance and chrominance noise removal the same set of sliders.
Figure 2: The Noise Ninja Interface (click for larger image).
Here, in Topaz, there is one slider that controls overall noise reduction, two that deal with luminance noise, and three for chrominance noise. That said, it does offer a bit more flexibility, but the downside is that the variables are not as intuitive. The software kept on surprising me during the two weeks of testing and usage. Every time I figured out that a particular setting would give me a certain result at a certain level, the next image where I tried it would prove me wrong.
One thing that Topaz DeNoise does not have that Noise Ninja has is a built-in unsharp mask function. Since images processed by Noise Ninja do benefit from dialing in a little bit of it, I will be using a weak setting for it in the review. Please keep that in mind when evaluating the test shots later.
That said, the interface itself is not terribly complicated. There are a minimum of tabs, and basically once you are happy with the sliders, just hit OK and go.
Testing and Results
Now, while noise does happen when you push up the ISO setting, it also can vary from situation to situation, and camera to camera. As a result, I decided to test it on my two current cameras, the Nikon D300 and the Panasonic GH1, in both day and night time settings.
Also, in order to make sure both of the noise reduction software packages are shown in their best light, the settings for each photo have been made to pick the best balance between noise reduction and detail preservation. It also means that I did not use the same uniform settings throughout the test. I also loaded Noise Ninja with the D300 profiles, which were available on the Noise Ninja website.
Photos were processed with Adobe Camera RAW, with the default 0 Luminance and 25 Chroma noise removal settings. Files were then saved to JPEGs at near-minimum compression levels.
The files used in the test are available for viewing. Use them to make up your mind on which aspect of detail vs. noise you wish to see. (30MB each)
Day Test Images
Night Test Images
Figure 3: Day time test crops. Full-sized images available above.
As you can see in the crops, Topaz DeNoise did a good job at removing luminance noise from the image, and manages to keep enough detail in the image. Note that the clumps of leaves in the shadows still have definition in them. Noise Ninja still has some luminance noise in the image, though it is a match here for Topaz DeNoise in the chrominance noise removal. Both the GH1 and D300 images show similar results for both plug-ins.
The Topaz DeNoise images may look softer, but the detail is still in the image. A little unsharp mask to increase the edge contrast will make the image look better. Topaz DeNoise gives the better results here.
Figure 4: Night time test crops. Full-sized images available above.
The night test, with much less light than the day test, taxes both cameras' sensors quite a bit more. Here, the differences are not quite so clear cut. Topaz on the whole with the right setting will produce cleaner files, at the expense of detail. Note that the image looks a little devoid of detail, especially in the shadows. Noise Ninja goes the other way; it is better able to preserve detail at the expense of more luminance noise. The GH1 images do lose more detail in the shadows for Noise Ninja, however.
At this point it really is a matter of preference, I feel. Do you need the cleaner look from Topaz DeNoise, or the slightly grainy but more detailed look from Noise Ninja? I think I'll call this a draw here.
Here comes the first of three complaints, and this is the minor one. Topaz DeNoise attempts to preserve detail when removing noise; unfortunately, it leaves some readout noise in place. What this means is that it can highlight the banding resulting from readout noise. Here is a sample (it is much clearer in the full test shot though):
Figure 5: Slight banding on this crop; if you see the full-sized image you'll notice plenty of such patterns all over the night sky.
Now, here comes my big complaint. Topaz DeNoise is SLOW. S-L-O-W. Here's how they stack up when removing noise from 12 megapixel files:
Noise Ninja: ~2 seconds
Topaz Denoise: ~45-50 seconds
This is on my main machine, with a 6GB of RAM and an Intel Core i7 920 overclocked to 3.0Ghz. Doing this review and all the iterations used when testing has been a pain, since waiting 45-50 seconds for each try gets old very fast after the fifth attempt or so.
I briefly did a search on Topaz DeNoise being slow and chanced on a couple of threads in both DPReview's forums as well as Topaz's own forums. It is acknowledged that Topaz DeNoise is not the fastest, but they are working on making it faster in the future. Currently though, if your machine is not "fast", like a notebook class Intel Core 2 Duo, expect 12 megapixel images to take 2-3 minutes to complete.
This would normally be a minor complaint, but here it is not. Topaz DeNoise ships with a number of presets, labelled as "JPEG Light", "RAW Strong", and so on. I discovered when playing with the presets that each set of settings will not give the same results from photo to photo (as I mentioned above). Before you go "doh, each photo is different", let me explain.
For example, in Noise Ninja, I normally set the noise reduction to a particular configuration of settings. They generally work well for 90% of the time, and the other 10% I just need to nudge the sliders a few levels one way or the other. Even so, the 10% is usually just a case of "little-less-than-expected"
In Topaz DeNoise, I found that different images require quite different settings. Generally as I prefer my noise reduction to be light, I tried using the "RAW Light" preset. Imagine my surprise than in some images (usually, but not always, the noisier ones), "RAW Light" felt like "RAW Does-nothing-whatsoever".
Slowness + Inconsistency = ?
Here is where inconsistency becomes a big problem. The implication of this is two-fold: Firstly, you have to do the trial-and-error way of testing quite a bit. This means spending time playing with the settings. Coupled with the slowness of the plug-in, it means that more time will be invested into getting the image look right.
Secondly, it also means you cannot work around the slow problem by batching, especially if it involves a large number of photos taken in a wider range of exposure and ISO settings.
I know I sound a bit negative here, but while good, and at times brilliant, Topaz DeNoise just is not the cure-all for everyone. While at times it can exceed the last generation of noise reduction software by a lot, there are also times it only barely bests the (older) competition. Coupled with the slowness and the inconsistency of the settings, it means the product just is not for everyone. My conclusion is thus two-fold.
For fine art photographers and anyone who only edits a few photos at a time while striving for perfection, this comes highly recommended. For wedding, event or any kind of photographers who deal with large volumes of photos, you may want to look at a faster solution that is almost as good. Also not recommended if you have an older camera and want to extend its life a bit more with better noise reduction solutions.
I look forward to the next version of Topaz DeNoise though; if they get some kind of GPU acceleration going and bring down the processing time to one-fifth (or even less) of what it is now, I think it will be an unreserved recommendation for any serious photographer.
Lee Yuan Sheng has been shooting (with cameras that is) for some 17 years, and never ceases to marvel at both what cameras can do now, and how people can complain about noise in this day and age. He is also a little old-fashioned in preferring a little grain in his images, so please hold off the emails saying the images shown are not silky smooth.
Do you enjoy using new hardware, software and accessories, then sharing your experience with others? Then join us on the Thoughts Media Review Team! We're looking for individuals who find it fun to test new gear and give their honest opinions about the experience. It's a volunteer role with some great perks. Interested? Then click here for more information.
Baka. Soku. Zan. - The justice behind the dysORDer.