5. Noise Performance and Image Imperfection Issues

Digital image noise might be obtrusive, but a virtually noiseless digital image can in fact be perceived as cold and sterile. Thus, not all kinds of noise are bad as far as photography is concerned. Just rewind your memory to the old glorious days of souped Tri-X when megagrain (not megapixel) was the vogue.

I think we quite easily accept some noise in the images when that noise manifests itself sufficiently randomly, in nature as well as in its spatial distribution. Visual patterns of noise, and in particular banding of the image, are far more objectionable. Unfortunately, the artifical kind of noise we have seen in the past with many DSLRs (from Nikon and their competitors as well) has earned the notion of digital noise a bad reputation in some photographic circles. In a manner of speaking, such noise detracts from the image content in the way that the earliest and poor digital records grate on the ear.

Underutilization of the dynamic range of the digital imager is a major putative candidate for noise generation. This occurs either by user error (underexposure) or by deficiencies in the metering system of the camera emphasizing highlights instead of shadows when scene contrast is high. I believe the matrix metering system of the D3 tries to prevent highlights to blow out and thereby sometimes allows the shadows to get too dark. The end result is an image with perfect highlights and shadows in which traces of colour noise can appear. The cure for this is simply to expose a little richer, or thinking about how you did shoot negative instead of slide film in the dark ages so many moons ago. When scene contrast is moderate however, the metering accuracy of D3 is uncannily spot-on all the time.

If you underexpose severely, banding tend to occur on most DSLRs, however, D3 is not much troubled in this area. The graininess of the image will increase by underexposure, of course, but the banding issue simply isn't there to any noticeable extent. The chrominance noise is quite randomly distributed. See the ISO test series presented later on this page to learn how the camera deals with increased noise.

Automatic white-balance is another less readily recognised culprit causing digital noise. We have become sloppy technicians preferring to click a button in software to get a "white" rendition instead of attaching a colour-balancing filter to the lens. Thus when shooting under incandescent lighting and setting the colour balance by automatic means, we are (but shouldn't be) surprised and annoyed by appearance of blue-channel noise, and don't consider the tremendous additional boost of blue colours needed to balance a blue-deficient light (incandescent light) as white. If you combine incandescent light sources and severe underexposure, the end result will show visible blue channel noise. Obviously you should expect such results, but still some people need to be reminded of facts before they start complaining about camera "flaws".

The D3 handles shooting strong point light sources extremely well. There is a similar manner as with film in how the light flares into its surroundings. I can only describe this behaviour as "graceful". If I had observed more of the sun in the testing period, I could have added into-the-sun shots that behaved nicely as well.

By the way, shooting into the sun is something you should do with great care, your eyes may not like the intense glare any more than does the imager in the camera. If you shoot into the sun (typically with wide-angle lenses), always remember to stop down to the smallest aperture value available on your lens. The only exception to this rule is when you shoot sunsets with telephoto lenses, a situation in which you normally wish to have the solar disc as large as possible, and hence should set the lens to be wide or nearly wide open. However, don't ever try to do this unless the sun is just about to set, or is enshrouded in a thick cloud layer

A final way of avoiding noise is to max out aperture (and/or shutter speed) before you dial in a higher ISO setting on the camera. I find I can achieve virtually noiseless images with the 35 mm f/1.4 Nikkor, one of my pet lenses for D3, at 200 ISO and f/1.4 @ 1/4 sec instead of setting it to f/2.8, use 1/15 sec and increase ISO to 800 - 1600. Now, with the D3, you might go way higher in the race for sheer ISO speed. More on this later.

The noise performance of the D3 can be assessed by lab-condition measurements, or evaluated by taking pictures under field conditions and look how the pictures appear to the eye. The latter approach is what I've done this time.

So, let's investigate various aspects of the image rendition of D3. The noise behaviour as a function of ISO setting is illustrated in the series of detail crops below. A lovely Moth Orchid was pressed into service as the appropriate shooting target for this exercise. I also availed myself of the excellent jpgs produced from the D3, in this case using the default settings.

Below is a series of 100% crops (meaning the crop is shown at actual pixel size so each pixel in the file is represented by a pixel on your monitor). I selected a crop which contains the entire gamut of shadow and high-light detail to demonstrate how image tonality is handled by the D3 under various ISO settings. I also added a brief analysis for each of the alternative ISO speeds you can get from this camera. Read these statements in the proper context, namely, as they would be manifested with a reasonably-sized print. Nit-pickers and pixel peeping toms always can find "flaws" when they look at what equals a huge print. I'm not concerned with such people.

The lowest setting is "Low 1", approximately 1 EV below ISO 200. Although not a calibrated and thus valid ISO setting, it is frequently assigned a value of ISO 100. In this case I had to open up the lens from the f/11 setting used elsewhere, since the low setting did not correspond exactly to "100 ISO". This slightly influences the rendition out-of-focus details in the background. It is also a warning that Low 1 really is an uncalibrated setting. since the D3 is digital not film, the lowest "ISO" won't give any better quality, instead exposure times are longer and the dynamic range becomes squished.

The base setting is ISO 200. This produced richly textured surfaces and excellent tonalities. Noise levels are extremely low so the image is rendered with a superbly delicate smoothness.

The 400 ISO is similar to 200 ISO. A full range of details and texture is preserved. The non-photographing pickers of nits may be able to observe the merest trace of noise, whilst the photographers would fire away and couldn't care less.

At 800 ISO, the slight traces of a film-like grain pattern start to creep into the lower tones and shadows. You would be very hard pressed indeed to see this in any print made from such a file.

At 1600 ISO, the first signs of noise are unequivocally present. Detail sharpness and colour rendition are not impacted to any degree, so high-quality prints can be made and nobody should be too concerned about "noise".

At 3200 ISO, some noise appears. and the tonalities take on a harsher impression. Detail sharpness and colour rendition are well preserved, but the very finest detail begins to be impacted by the noise..

At 6400 ISO, the highest calibrated ISO setting of the D3, noise is plainly visible and the tonalities take on a harsher impression. Yet the image structure is coherent and is not troubled by the ugly blotchily distributed colour artificats seen on many DLSRs. Detail sharpness and colour rendition still are well preserved, but the finer detail are degrading. Nice prints could still be made from these files, but of course, now post-processing actions become more significant in how the final printed image would appear.

At the Hi 1 setting, approximately 12800 ISO, we are into the boosted range of ISO settings. The increase in noise is clearly visible, but the image coherence is still really good and for many subjects under poor light conditons, quite respectable prints are still possible. But with the associated loss of fine detail and increased graininess, post-processing has to thread lighly in order to preserve the image content and not exacerbate the noise pattern.

This is Hi 2 setting, approximately 25600 ISO, and obviously intended mainly as the last-ditch resort to get an image under absolutely appalling low light conditions.we are into the boosted range of ISO settings. The increase in noise and concomitant loss of image detail is plainly visible. Some "hot" or "stuck" pixels appear here as well. The noise pattern is slowly approaching a non-filmlike and blotchy appearance, making the post processing of such files more of a challenge than some users may find comfortable. You can, however, convert the files into black/white prints with good results.

All images © Bjørn Rørslett/NN

Chromatic aberration (CA): An issue gone away with the D3?

What is immediately apparent is the low amounts of chromatic aberrations (CA) seen on D3 images. The LCD display on the camera really shows you the output from the powerful EXPEED imaging engine of the camera, and the result is carried over to the in-camera generated jpgs as well. There is no doubt EXPEED removes a vast fraction of any CA present. Moreover, EXPEED also greatly reduces the vignetting of lenses. Some lenses will tax the EXPEED engine outside its workable limits so the camera responds by flashing "Err", however, and sometimes no picture will be taken. I noticed this strange behaviour with my venerable 15 mm f/5.6 Nikkor-QDC. The 18 mm f/3.5 Nikkor exhibited greatly exaggerated vignetting set to f/3.5, so only the very centre of the frame was bright and the rest receded into darkness. Stopped down to f/11 the extreme vignetting disappeared and sharp images were obtained. To contrast these discomforting observations, my tilt/shift lenses produced crisp and clear pictures across the entire frame with no traces of the CA illness and loss of sharpness so prevalent with these lenses on a D2X.

The obvious question now is how the initial processing done in-camera is applied to the output formats. We already know the jpgs straight off the camera have remarkably low CA levels and there is greatly reduced corner light fall-off with shorter lenses (but see the caveat of the previous paragraph). It's easy to verify that TIF files delivered off the D3 show the same behaviour. So, what is the situation with the RAW (NEF) output?

Nikon D3: In-camera jpgs vs NEFs

Centre Corner Origin In-camera jpg NEF by Capture NX 1.3, no CA removal NEF by Capture NX1.3, CA removal NEF by Adobe Lightroom, CA removal

These test shots are taken with the 50-135 mm f/3.5 Zoom-Nikkor lens at f/8, 135 mm setting at its near limit where this lens shows clear signs of chromatic aberration (CA). Also, there is pincushion distortion as evidenced on these 100% crops from the centre and corner of the frame. The crops are from the same NEF and its associated jpg (fine) at default settings in-camera. Detail magnification m = 0.117 (1:8.5)

From my test shooting I conclude that the NEFs carry with them the lower degree of vignetting also seen on the jpgs straight off the D3. Either this is an unprecedented virtue of the EXPEED's wizardry or a result of the new sensor array design. The alternatives are hard to tell apart because we can't circumvent the internal processing before image data is written to the CF card. Not that it matters anyway as long as this works so well. But what about CA issues? Here the story is different. CA as such is not removed from the NEF at all, but can be virtually eliminated in the subsequent processing stage by by computer software. As evident from the above comparison panel, third-party software can do almost as well as Nikon's own Capture NX. Another remarkable observation is the high quality of in-camera generated jpgs.

Nikon D3 Reviewed