Noise

2006-08-01 • SEND FEEDBACK •
Related: digital artifacts, fairy dust aka hot/stuck pixels, moiré, Nikon, Nikon APS_C, Nikon D200, Nikon D2X, Nikon DSLR, noise, post processing, shutter

Introduction

When applied to a digital image, the term “noise” (an auditory term!), refers to random or semi-random variation in individual pixels.  It can manifest itself as either luminance noise (variations in brightness) or chroma noise (colored speckles).  In particularly bad cases, it can become blotchy and/or streaking/banding/striping occurs.

Test parameters

• All noise reduction was turned off unless otherwise specified.
• At each ISO value, the same shutter speed and aperture was used with the same lens for each cameras.
• Raw processing was gray-balanced in Nikon Capture using the 4th gray patch from the left on the MacBeth ColorChecker card shown above.
• Temperature was approximately 50° F.  Digital noise is reduced at lower temperatures, so higher temperatures might show increased noise.

All images were shot in RAW format, then processed without sharpening, using the defaults as described on the Raw File Processing page.

Making a valid comparison

Making a valid comparison is more complicated than it might seem:

• The white balance should be accurately set using a neutral gray card.
• Actual and true ISO potentially vary slightly between cameras, so the same shutter speed and aperture might not produce the same brightness; this can impact the perception of noise. However, the D200 and D2x appear to have the same ISO response.
• The cameras potentially have a different dynamic range, rendering the tonal range from pure black to pure white somewhat differently.
• The tonal curve affects the visibility of noise, and it might be difficult to match the tonal curve between cameras.
• The choice of raw converters, and the tonal curves they use affects the perception of noise.
• There is a subjective element to noise; pure measurements do not tell the whole story.

Because of these issues, minor differences become questionable as a basis for firm conclusions.

Fortunately, some of these variables can be controlled: Nikon Capture was used for raw-file conversion.   The identical scene with the same lighting, lens and tripod position were used to make the images for each camera, with a slight adjustment of camera angle to include the same portion of the image (the camera heights are an inch or so different, so a slight adjustment is required to frame the subject the same way).

Degree of enlargement

A higher-resolution camera’s actual-pixels crop which shows the same level of noise as a lower-resolution camera’s actual-pixels crop is actually a better performer because the resulting image will need less magnification when making a print.

The D2X resolves 4288 X 2848 pixels horizontally, as compared with 3872 X 2592 for the D200.   A D200 image will thus require 10.7% more enlargement horizontally than a D2X image. Any noise in the original image is thus enlarged by 10.7% as well.

For this reason, evaluation of noise must be done at the same resolution, thus requiring the upsampling of the D200 image.

Large crop ISO series table: 100, 200, 400, 800, 1600, 3200

Consistent with the degree of enlargement discussion above, the large-crop images shown below have been upsized to match the D2X horizontal resolution of 4288 pixels.  JPEG compression was used on each resulting crop (Photoshop “Maximum” quality).

The eager reader might be especially interested in seeing what a “real” picture looks like at each ISO value with each camera, rather than the uninteresting Macbeth color checker patches shown later on this page.  The table below contains mini images, each of which links to a much larger crop.

Click on each image in the table below to see a large version of each, or see all the images together in one huge table.

	ISO 100	ISO 200	ISO 400	ISO 800	ISO 1600	ISO 3200
D200
D2X

The two ISO 3200 crops are shown below.   Note how the D200 image color rendition remains consistent with the lower ISO images (see the table). The D2X image shows a desaturated look with a color shift.  (Also apparent is the higher resolution of the D2X).

D200 ISO 3200, upsized to match D2X resolution

D2X ISO 3200, actual pixels

The D200 does offer slightly better control of noise, though this is hard to detect until ISO 1600.  But what may be more interesting is the D200’s more consistent color at ISO 3200; the D2X at ISO 3200 produces less saturated and dull color.  Look at the large crops to see this.

It is also interesting to note the reduced moire with the D200 in the cloth below the doll’s chin (see below).  Moire is an interaction between the pixel pitch of the sensor and repeated patterns in the subject; in the case, the D200 results are not moire-free, but do show much less of it than the D2X.  This could also be due to the anti-aliasing filter employed in the D200, but in this case it’s probably due to the slightly lower resolution of the D200.

Moire
D200	D2X

OWC ThunderBay 4
High capacity, high-performance fault-tolerant storage for photography and video.

Non-RAID or RAID-0/1/4/5/10.
Capacities up to 72 Terabytes!

More about OWC Thunderbay 4...

Hard drives or SSD.

Gray-scale noise: ISO 100, 200, 400, 800, 1600, 3200

Consistent with the degree of enlargement discussion above, the D200 crops that follow have been resized to match D2X resolution.   This series shows a section of the Stouffer R2110 step wedge.  The slight yellow cast of the D2X crops might be due to a slight change in the shooting angle (due to the different heights of the camera), but that is not certain;  this is explored on the Color page.

Nikon D200 step-wedge noise Tone=Low, no sharpening
100	200	400	800	1600	3200

Nikon D2X step-wedge noise Tone=Low, no sharpening
100	200	400	800	1600	3200

Now let’s compare the high ISO results ( ISO 400/800/1600/3200).  The D200 has approximately 23% larger pixels (area) and thus we should expect to see lower noise as compared with the D2X.   Inspect the side-by-side crops, then read on for conclusions.

Step-wedge noise
D200 ISO 400	D2X ISO 400		D200 ISO 800	D2X ISO 800

At ISO 400, distinguishing the noise levels between cameras is difficult; they are essentially identical.  At ISO 800, the D2X hints at a bit more chroma (color) noise than the D200.

D200 vs D2x step-wedge noise
D200 ISO 1600	D2X ISO 1600		D200 ISO 3200	D2X ISO 3200

At ISO 1600, the cameras are still so close that distinguishing them is difficult.  At ISO 3200, the D2X does display more chroma (color) noise, though the D200 noise pattern has a “chunkier” look to it, perhaps because of upsizing the D200 image. 

Gray patch: ISO 100, 200, 400, 800, 1600, 3200

The crops below have not been resized; they are as-processed actual pixels crops from the dark-gray (next to the black patch) on the Macbeth ColorChecker chart. The luminance of this patch is about 35 in Lab color space (RGB value depends on the RGB color space).   As seen in the step-wedge comparison, there does seem to be more color noise in the D2X results.

ISO	D200	D2X
100
200
400
800
1600
3200

Below is ISO 3200 again, but this time with the D200 upsized to match D2X resolution.  The D2X has more color noise than the D200.

D200 Upsized to D2X resolution	D2X actual-pixels

Noise reduction in Nikon Capture

Nikon Capture offers noise reduction for raw (NEF) files.  Please see Nikon Capture NX2 at diglloyd.com.

Please note that the camera settings when shooting RAW (NEF) are irrelevant for “High ISO Noise Reduction”; their only effect is to default the noise reduction setting in Nikon Capture to the chosen value.

Hot pixels

Some time after I performed these tests, I noticed a persistent “hot pixel” with the D200, always in the same location, and consistently appearing at ISO 800. Further investigation showed that the hot pixel actually appears even at ISO 100, though it is not visible without adjusting the brightness level.

Nikon D200 “hot pixel”
100	200	400	800	1600	3200

Processing the images with a 2 stop “push” (as might be done for an underexposed frame within the raw-processing program and/or with Photoshop “Levels”) shows that that the hot pixel is indeed present at all ISO values, and is larger than just one pixel (probably due to Bayer-pattern interpolation).

Nikon D200 “hot pixel” “pushed” 2 stops
100	200	400	800	1600	3200

Below is the ISO 100 2-stop push shown at 300% actual pixels. The actual number of affected pixels is a 3 X 3 grid at ISO 100, worsening at higher ISO values:

Hot pixel enlarged, “pushed” 2 stops 300% actual pixels for ease of viewing
100	200	400

While this red spot can easily be removed in an image editor, the fact that it appears on every frame is a nuisance. One or two other hot pixels do appear at ISO 1600, but are then obliterated by noise at ISO 3200.

I have not observed such a problem with the D2X. For that matter, I don’t even know if Nikon considers such a hot pixel a problem. It is a minor problem for high ISO shooting, but whether it is worth sending an otherwise functional D200 into Nikon Service is questionable.

Conclusions

The D200 offers lower noise than the D2X, though its advantage becomes noticeable only at ISO 1600 and not obvious until ISO 3200.  More important for some, ISO 3200 images have better color with the D200 than with the D2X.   This may be due to the apparently higher chroma (color) noise in the D2X, which affects the overall color rendition.  Those who regularly use ISO 3200 may be advised to choose the D200 over the D2X.

The character of the noise might be more important than the amount, but the D200 and D2X exhibit noise that looks very similar in character.  Users of early digital SLRs complained of streaking, or other pattern-noise problems, which looks unnatural as compared with uniformly-distributed film grain.  Neither of the cameras displayed any streaking in this example.

The appearance and nature of noise depends on many factors, including subject matter, ISO, shutter speed, internal camera temperature, raw conversion parameters, sharpening, etc.  There might be circumstances where the amount of noise is of less important than its other characteristics, but this test did not detect such differences.

Quite rationally, many users will ignore the issue of noise, and simply use the lowest ISO required to ensure a sharp picture!  The noise we’ve seen is far better than the equivalent film grain.   With a compelling photo, ISO values up to 1600 are not going to ruin the photo.  And in modest-sized prints, it's not even clear that the noise would be a prominent factor.