Introduction and Summary Findings
To judge from the various threads in discussion forums and web reports with performance tests, the arrival of the latest crop of DSLRs has generated a renewed flurry of interest in the subject of two big factors determining image quality: noise and resolution. It’s natural – one wonders whether new technology really allows, say, cramming 21 million photosites within a 24*36mm rectangle at tolerable noise levels, or whether the engineers cleverly designed processing algorithms which “simply” smooth out the noise at the expense of image detail, or whether there’s an optimum combination of both things happening. We don’t know the “whys and wherefores” within the camera, but we can look systematically at the results. Noise and detail are the flip sides of the coin because noise can obscure detail, as can noise-reducing firmware and software. Lens aperture and exposure technique also affect apparent resolution, therefore one needs to examine these as well in relation to noise and resolution.
Money really helps concentrate the mind, so after plunking-down 8000 bucks for a 1DsMarkIII (1Ds3), testing it at ISO 200 with the High Tone Priority1 (HTP) option enabled and seeing slightly “grittier” results in the three-quarter tones than expected for this ISO, it invited exploring the related issues of noise and detail more generally. As for HTP, we’ll dispense with that first –the manual says that by enabling HTP “noise in the shadow areas may be slightly more than usual”. So this is turned off (with confidence we can manage tonality in Camera Raw [ACR]); all the following discussion assumes HTP is disabled.
Here are the five conclusions emerging from this exploration:
(1) The 1Ds3 and the Nikon D3 are superb performers in respect of both noise and resolution, judged from Super A3 size real-world photographs2 printed on high DMax media in an Epson 3800.
(2) That said, correct digital exposure technique (Expose to the Right – ETTR) is a must for achieving optimal results.
(3) The noise you see on a display may or may not be a reliable indicator of noise you will see in a print, depending on the size of the print.
(4) Beyond a certain threshold in the lower noise range, the noise measurements from noise reduction software is a useful indicator of noise you will see in a print.
(5) Using the Canon 24~105 f/4 L lens, resolution visibly deteriorates relative to the “sweet range” when the aperture is narrowed beyond f/11, but sharpening mitigates it.
If that’s all you want to know, you may click out. If you are interested in the details, please read on.
1 HTP, according to the manual “Improves the highlight detail……….the gradation between the grays and
the highlights becomes smoother.”
2 A real-world photograph is a picture of anything other than test charts and resolution diagrams.
Before testing anything, one must be clear about what one is testing for (noise and detail), how to do it and how to evaluate what one has done. The latter is particularly difficult because apart from the world of test-charts and instrument-based measurements, judgments about noise and detail as they affect the appearance of real world photographic prints (our interest) are rather qualitative, and depend on image size and viewing distance. One can be aided by the measurements created with image-specific noise profiles made in applications such as Noise Ninja and Noiseware3. As for detail, we have no interest in photographing resolution charts. We prefer to photograph subjects with fine detail (near and far) and look at how well camera and lens, (with or without sharpening in ACR and/or Photoshop), render them. All the images discussed in this paper are made with the Canon 1Ds3, the Canon 1Ds (for comparison) using the same Canon 24~105mm f/4 L lens, and the Nikon D3 using the Zoom Nikkor VR 24~120 mm lens.
If the end product is usually a print, it can be insufficient to only evaluate noise and resolution on a display4. We find that the display image, probably because it’s transmitted light, seems to reveal more stuff that may look like artifacts (especially at magnifications from 50% upward) than visible in smaller prints, though there is less of a disconnect between the display image and the print when viewing larger prints; as for revealing image detail, an Epson 3800 most likely out-resolves most of our display.
Our test-bed is the Hillcrest Community Centre because the scene has the combination of light, shadow, fine detail and open sky (around 2:30~3:30 PM on those rare comfortable, sunny, cloudless Toronto days) needed for this work.
Figure 1. The Hillcrest Community Centre, Toronto
(Canon 1Ds Mk3; f/6.3 ISO 200 @ 45mm)
The testing was in four phases. Phase 1 is to see the Airy disc effect5, the purpose being to determine whether the choice of aperture could have a significant impact on our appreciation of how noise affects image clarity; in Phase 2, we explore differences between correct exposure seen with the in-camera histogram, versus the Camera Raw histogram6; in Phase 3, we assess the impact of ISO and exposure technique on the appearance of noise in images, and in Phase 4 we deal with noise reduction, sharpening and prints.
3 Even with these applications the correlation of their measurements with one’s visual observations is not
always consistent for a number of reasons, but the measurements help to make indicative comparisons.
4 ….even high resolution displays of the kind we are using: Lacie 321, Apple Cinema.
5 Named after George Airy, this is the effect of decreasing resolution with narrowing lens aperture beyond
the lens “sweet range” due to the diffraction of light within the lens.