4. High ISO
The next section in this RAW converters comparison is to examine how well the applications can process images taken at high ISO values.
Resolving detail in images does not generally just involve the manipulation of sharpening and micro contrast, but also involves the interaction of these tools with those for managing digital noise reduction.
The interplay of these tools is particularly complex and getting optimal results is very difficult. If you gave an image to 10 people to process with these tools then you would almost certainly get 10 very different results. Noise reduction tools and sharpening tools work against each other and striking the right balance is very difficult. If too much noise removal is applied, details are smeared and sharpening only makes this more obvious, giving a crunchy blocky appearance; if too little noise is removed then sharpening sharpens the noise itself, giving a crunchy speckled appearance instead.
Digital noise tends to show up in two main situations for the landscape photographer, in situations where using a higher ISO setting was necessitated due to low light levels, and in situations where shadows have to be opened up due either to underexposing a shot or because of a very high dynamic range in the capture. We will turn to these latter situations in subsequent sections; here we will stick to noise generated by turning up the image sensor sensitivity in the camera itself.
I very rarely shoot above base ISO due to my noise aversion so I don’t have many images in my catalogue to choose from here. The one I have chosen is one I took at dusk, shortly after I acquired the E-M1, specifically so that I could check the camera’s behaviour at higher ISO settings. I have selected it from several I took at the time because the details (not the sky!) are well exposed with no blown highlights or shadows. It was taken with the 12-40mm at 17mm, ISO1600, 1/30s and f/7.1.
The whole image looks like this (processed on Lightroom defaults):
Here is a 1:1 crop of the image of the rocks at the bottom (no sharpening or noise reduction): [Apologies to Retina display users, for whom it will look even worse.]
My aim in processing this image is to find the near impossible balance that almost entirely removes obvious noise whilst still trying to resolve as much detail as possible. In normal processing of this image most people would leave in far more noise than I have done here (including me), especially on this crop, but that is not the point of the exercise. If I chose a part of the image with low frequency detail then we would not be able to learn how well details can be retained when the denoising algorithms are applied.
The RAW Conversions
Settings & Notes
Sharpening – Amount: 45; Radius: 0.8; Detail: 60; Masking: 82
Noise Reduction – Luminance: 40; Detail: 70; Contrast: 50; Color: 25; Detail: 50; Smoothness: 50
Notes: As I stated in the introduction to the article I struggle with the combination of Lightroom’s noise reduction and sharpening tools. This really is my best effort after half an hour of work.
Noise Reduction – Radius 1.6; Edge Detail: 3.0
Edge Sharpening – Intensity: 0.6; Edges: 0.25; Falloff: 1.0
Notes: No distinction is made between colour and luminance noise reduction. I am suspicious that colour noise reduction is done automatically with little user control.
Capture One Pro
Clarity – Algorithm: Neutral; Clarity: 2; Structure: 7
Sharpening – Amount: 80; Radius: 0.8; Threshold: 1.0
Noise Reduction – Luminance: 25; Color: 50; Single Pixel: 0
Notes: This is a ISO1600 image yet the values being applied here are still less than half the defaults Capture One Pro applies to every image no matter the ISO it was shot at.
Local Contrast – Strength: 15; Sensitivity: 6
Sharpening – Amount: 50; Sensitivity: 8
Raw Noise Removal – Raw noise: 20; Threshold: 10
Perfectly Clear Noise Removal – Strength: 20; Detail: 10
Notes: Very difficult to figure out the relationship and interaction between the two noise reduction tools; the documentation supplies little enlightenment.
DXO Optics Pro
Lens Softness – Global: 0.5; Details: 80; Bokeh: 50
PRIME Noise Reduction – Luminance: 20; Chrominance: 50; Low Freq: 0; Dead Pixels: 24
Notes: PRIME noise reduction is beyond painful to use; even the small preview window takes an age to update, and a full render, taking several minutes, is really the only way to assess an image. On this image the default setting of 100% low frequency noise reduction was disastrous and it had to be disabled altogether.
Difference of Gaussians Sharpening – Noise Reduction Radius 0.4; Sharpening Radius 0.7; Amount: 200
Noise Reduction – Adaptive Early Stage: 5; ChromaLogic: 2; ChromAdaptive: 5; Luminance: 10
Notes: The Difference of Gaussians sharpening algorithm is specifically intended for noisy images.
Sharpening – Strength: 50; Radius: 0.8; Noise Masking: 100
Noise (Noise Ninja 4) – Luminance Smoothing: 30; Residual Noise & Detail: 90; Color Strength: 50; Defringe: 0
Notes: Possibly even more painful to use than DXO PRIME, even when attempting to use the Turbo setting as a preview of the high quality version, as there is no preview window and each change to the image one luminance noise reduction is enabled leads to very lengthy processing times. The results are so different to other noise reduction algorithms that it is hard to assess them without more experience with the software.
RAW Photo Processor
1.00eV Compressed Exposure increase
Topaz Denoise – Strength: .18; Shadow: 0; Highlight: -.72; Red: -.54; Blue: -.5; Clean Color: .13; Black Level: 1.00; Recover Detail: 0.15; Reduce Blur: 0.21; Add Grain: 0.18
Focus Magic: 1 Pixel
Notes: Fairly high settings in Topaz Denoise and it struggled with the shadow areas. Focus Magic does not like high ISO images much so it was only run at 1 pixel rather than the 2 pixel default.
For the first time in this RAW converters comparison clear differences emerge in the results.
As noted in the introduction my personal workflow has long involved turning off sharpening in Lightroom and going to Photoshop to use Topaz Denoise to deal with any luminance noise reduction that the image requires. The results here give a pretty good indication of why I have been doing this. The image as processed with Topaz Denoise and Focus Magic on top of the RAW Photo Processor rendition has probably the sharpest detail of all.
Of the pure RAW conversions Irident Developer’s specialist sharpening and noise reduction tools deliver the goods here with an incredible amount of detail resolved in such a high ISO file.
Just behind, with a slightly more painterly look, is Capture One Pro’s rendition, and just behind that is DXO Optics Pro. DXO delivers almost three dimensional detail in the mussels clinging to the foreground rock in the larger image but then smooths over and loses most of the detail in the larger rock underneath it. Another significant difference here is processing time. Capture One Pro could render this image in less than a minute whereas DXO’s much vaunted PRIME noise reduction led to it taking took nearly four minutes on my machine.
I am not sure how to view Photo Ninja’s effort. In a printed file it might well turn out that Photo Ninja delivers a very pleasing result but here, at 1:1, I find it hard to look at. The larger image almost makes me think I have double vision. I may not be using Photo Ninja’s noise tools to the best of their ability (not least because they are nearly as painfully time consuming to use as DXO’s) and I am unable to remove the very fine grain that it adds to the image. That fine grain works very well in appearing to give the image greater high frequency detail than it really has. However, to me, in this image, it looks strange.
Aperture looks a bit painterly but its rendition is vastly preferable to both Lightroom and Aftershot Pro. The latter is too soft and the former too painterly by far. I spent longer with Lightroom’s multiple and complexly interacting sliders than with any other image here yet it only just escapes being the worst rendition of all. Maybe a more skilled operator than me could deliver a better job but I would hope that a tool would facilitate me getting a decent result in five minutes. If I have to spend half an hour battling with it then it is not, in my opinion, a good tool.
It is not simply the case that the tools that spend the longest applying noise reduction are the most effective. DXO and Photo Ninja’s highest quality algorithms take far longer to run than Topaz Denoise, Irident Developer and Capture One Pro. However, at the other end of the scale, the algorithms in Aperture, Lightroom and Aftershot Pro are practically instantaneous, and this in all likelihood does say something about their quality. I frequently read people saying that they try to avoid luminance noise reduction because it smooths detail and gives surfaces in images a plasticky look; I’ll stick my neck out here and say that if I was only familiar with Lightroom or Aperture then I would likely say exactly the same thing. Fortunately, better tools are available.
We’ve made some assessment of noise reduction capabilities here, as applied to high ISO images; we’ll also see it in action in the next section of our RAW converters comparison when it is applied in shadow recovery processing.