White balance is the nearest digital equivalent to “daylight film stock” or “tungsten film stock.” The actual response of the image sensor to light is designed and fixed at the point of manufacturing, so adjustments for white balance are done electronically. This does mean that some white balance settings will require less amplification of the individual RGB channels than others, and so some will be less noisy than others.
Usually settings that require the least blue-channel amplification will be quietest, since blue is the color that electronic imagers see least well — many CMOS imagers are described as having a “native tungsten balance” due to their greater sensitivity to the red end of the spectrum. On modern cameras with a very low noise floor this entire concern may be effectively moot. Even so, it’s still quite possible to use optical filtration before white balancing, if there's enough light to waste, in order to move the colorimetry of the image closer to the sensor’s sweet spot.
Unlike RAW images, the H.264/MJPEG video will not hold up to a lot of image manipulation in post. For many photographers, this takes getting used to because it means having to get the image right in camera, with very little room for error. Correctly white balancing the camera is crucial for this reason. Most of the time, manual white balance will produce better results than the camera's auto white balance feature. The simplest way to do this is to aim the camera at a white surface, shoot a picture, tell the camera to reference the white from that picture, and then set the white balance mode to “custom.” For many situations this works fine, but sometimes finding a proper white surface wastes time and may produce an inaccurate balance.
If a slightly warmer or cooler look is preferred, white balance can still be set to an accurate white, and then the look can be manipulated in the camera. Users can effectively set the camera to lean warmer or cooler in relation to the white balance that’s been set. This negates the need for special warming cards and is generally more accurate because there’s a finer level of control over the desired look.
HTP is a Canon-originated term that describes a formalization of something that has often been done for electronic imaging before: quite simply, underexpose and recover brightness in postproduction, compressing the highlights in the process. The purpose of it is to retain highlight detail and smooth highlight rolloff, while maintaining the exposure of the rest of the image.
The approach used by Canon is a good generalization of the technique. The camera will intentionally underexpose the image by what appears to be one stop, simply by selecting an ISO setting one less than the one that's actually selected. For this reason, HTP doesn’t work below ISO 200.
Ordinarily, this would result in a one-stop underexposure of the image. However, selecting HTP causes the camera to apply gain to the signal, outside of the usual context of picture styles. This normalizes the resulting brightness of the image except in the highlights, where the amount of gain is gradually rolled off. This has the effect of creating a picture as it would have appeared had it been shot normally, but with a much smoother transition between “very bright” and “white”.
This technique can be simulated in any camera that accepts an uploaded luminance curve similar to Canon’s Picture Style editor, and it’s common practice to define such a curve that reduces highlights and amplifies shadows in order to make the best of available dynamic range. Really extreme attempts at this may mean the camera’s built-in light metering encourages an exposure that doesn’t produce the right output, but that’s what tests are for.
Whether this technique actually increases the real dynamic range of the imager is largely a semantic discussion. Dynamic range is the difference between “black” and “white”, where “white” is easy to define as the point the sensor hits saturation. “Black” is a tougher call; there is always thermal noise even on a sensor that has seen no light at all, so it becomes clear that the actual dynamic range is controlled by how much noise is tolerable. By underexposing, then amplifying, noise is increased. The acceptability of this is up to the individual.
There are generally two reasons to take sample images, and they demand different approaches.
The first is simply to evaluate exposure under the available light. This can’t be done qualitatively on the camera’s LCD, and it certainly can’t be done in the field in an uncontrolled monitoring environment. Since the intention is to shoot video, and because the camera will handle video somewhat differently than it will handle a raw still, it’s a good idea to shoot short video clips as well as any JPEG or RAW stills. This is especially important when shooting subjects which may cause moiré. This problem is visible on video but absent on stills. A proper assessment of moiré will require shooting video at various distances and with the camera at various angles to the subject, as the issue is predicated on the relationship between the distribution of pixels on the sensor and the image pattern falling on them.
The second and perhaps more exacting reason to shoot samples, is in order to define picture styles (to borrow Canon’s term) that will be used on the final material. This process is discussed elsewhere, but it’s important to shoot raw stills for this purpose. A raw still will give the picture style editor exactly the same information in terms of pixel values that the camera has when it processes the video image (though stills never show moiré).
In an ideal world, the relationship between these three measures is a simple one: one f-stop represents a doubling of light intensity, and therefore has the same effect as doubling film speed. The same scene can be shot at f/4 on 100 ASA stock or at f/5.6 on 200 ASA stock, and the results will be closely related to the digital equivalent of boosting ISO: the faster stock will have more grain and the faster sensor more noise, but the tradeoff is similar.
The video world has traditionally used the concept of gain to represent electronic increases in sensitivity, probably because it originally used analog processing amplifiers to modify images. DSLRs work in exactly the same way: they use gain — simple signal amplification — and relate it in terms of ISO that are familiar to photographers. Gain is measured in the logarithmic decibel (dB) scale using the 20-log rule, and since: 10 (6/20) = 1.995 (which roughly equals two), a gain of 6dB will double the output, equivalent to opening an Aperture one f-stop. Assuming a scene is being shot at f/8 on a camera with a sensitivity of ISO 100, and at 1/48th shutter speed (a 180-degree shutter at 24 fps in film terms,) correct exposure is achieved. ISO, gain and aperture then play against one another like this:
|100||F/8||6dB||One stop over|
|100||F/8||12dB||Two stops over|
|100||F/11||0dB||One stop over|
|100||F/11||12dB||One stop over|
|100||F/16||6dB||One stop over|
|100||F/16||12dB||Two stops over|
|200||F/8||0dB||One stop over|
|200||F/8||6dB||Two stops over|
|200||F/8||12dB||Three stops over|
|200||F/11||6dB||One stop over|
|200||F/11||12dB||Two stops over|
|200||F/16||0dB||One stop over|
|200||F/16||12dB||One stop over|
No professional videographer would be caught dead using automatic exposure. The process of constant evaluation and adjustment used by most cameras—including video-capable DSLRs—will in almost all circumstances produce distracting shifts in brightness as the subject changes. These are much more obvious in video than stills. Actually making a DSLR lock its exposure, however, can be surprisingly complicated. For instance, the Nikon D90 can be programmed (under Custom Settings in the Controls menu) to assign the AE/AF lock button to “hold”. This will lock exposure, but must be pressed after the recording of a video clip has started.
There are three ways that any camera can control exposure: by altering gain (in effect changing the sensitivity); by altering shutter speed; or by altering Aperture. A modern DSLR with servo aperture control lenses has absolute control over all of these and releases that control to the user on its own terms. The Canon 5D, in its original form, had this problem, and would actually alter its setup when the record button was pressed. A user could set up a combination of ISO, shutter speed and aperture while in live view mode, but the camera was fundamentally locked into Program AE while recording.
Since Canon’s firmware update 1.1.0 for the 5D, this has not been a problem, and the camera now obeys the “M” (Manual) dial setting in video mode. It’s therefore worth checking that a rental 5D has had this update applied, and that any other camera that will be used is well-behaved in this regard.
One of the earliest solutions to the 5D’s recalcitrance was to use fully manual lenses, often mount-adapted motion picture types that had only manual aperture rings. This removed that option from the camera’s control. The camera could also be forced to open the aperture on a servo lens by using neutral density filters, but both of these options are potentially dangerous. A camera starved of light may start to boost ISO, effectively using gain, to achieve a viewable image. The noise increase caused by this is easily missed on the camera’s tiny LCD screen.
The application of a traditional light meter to digital photography, whether stills or HDSLR video, can be tricky. A key part of the setup of any light meter is the sensitivity of the medium, which is usually an ISO setting. Digital cameras merely simulate ISO settings.
Techniques and standards that photographers apply to film will be at least somewhat different when shooting digitally, and there is some question over the usefulness of a light meter. There is a marked tendency to overexpose, even before considering the film-oriented rule of thumb that overexposure is better than underexposure. The exact opposite can be true on DSLRs, especially those such as the 5D Mark II and 1Ds Mark IV, which have vanishingly low noise.
In a world where the DSLR has histograms and is almost as portable as a light meter, even on a location reconnaissance it is often possible to take the camera along to evaluate available light. People who prefer to use a light meter should expose with these considerations in mind, generally erring on the side of underexposure and checking—or having an assistant check—the histogram to verify the correctness of the approach.
The histogram is widely regarded as the digital equivalent of a light meter. There are various types that show either overall luminance or, perhaps more usefully, the levels of individual RGB channels. They all do more or less the same job, with shadows at the left, highlights at the right:
A peak at fifty percent, for instance, would indicate that a large amount of the image has 50% brightness. Increasing exposure (either by increasing equivalent ISO, shutter speed, or decreasing Aperture) moves that peak to the right until it hits maximum brightness and the subject's detail is lost to clipping; decreasing exposure has the opposite effect. This is a great help when employing photographic concepts such as the zone system. This, however, takes some getting used to because it’s not necessarily obvious which part of the image is provoking which peak on the histogram.
A video waveform monitor, such as might be seen in a nonlinear editor, provides somewhat equivalent information and looks similar, but is not the same. A histogram has brightness on its horizontal axis and population—that is, the amount of the image that has that luminance—on the vertical axis. Shoot a perfect 50% gray card, and it will show up as a single peak on a histogram at 50%.
A video waveform monitor, on the other hand, is based on the video scanning concept of an image made up of a series of horizontal rows; it represents every pixel row of the image in its display. Its horizontal axis represents this composite of the rows of the image, and the vertical axis represents brightness. It can also be used to evaluate luminance, but a 50% gray card looks like a horizontal line, indicating that the entire width of the image at all vertical positions has 50% brightness.
Most DSLRs provide a histogram display, but it may obscure the LCD viewfinder image; worse, most of them will not display a histogram on the LCD while providing a video monitoring output via HDMI. This can be worked around by converting the HDMI output to HD-SDI, using something like the Aja HA5 converter; then feed the results into a test-and-measurement system such as the Astro DM monitor series, which will provide various analytical displays as well as picture monitoring. This is an expensive option, however; almost certainly more expensive than the camera, and most people will simply flick the camera in and out of histogram display.