Skip to main content

Colour control in the camera

There are two important issues in the camera: taking characteristics and transfer characteristics.

Taking characteristics

Taking characteristics are the spectral responses of the camera sensors, the wavelength-by-wavelength colour response. The user has no control over them because they are a feature at the time of manufacture. Ideally, they should be the colour-matching functions of the system primaries, mathematically defined across the visible spectrum (380nm to 760nm). (See Figure 1.)

In practice, cameras can only produce positive outputs to light stimulus. Figure 2 shows the response of a typical camera.

The chromaticity diagram in Figure 3 shows the colour reproduction of a GretagMacbeth ColorChecker chart. Each colour is represented by a blob with a vector pointing to the reproduced colour. Clearly, not very good. Although the colour errors all seem to be in saturation only, the vectors all point towards the white point, marked as an open circle.

Fortunately, the negative-going responses can be emulated in the camera by using an optimized linear matrix. This does far more than just change saturation; it corrects hue and lightness errors as well, sharing the errors within practical limits.

HDTV cameras usually offer several matrices. The best choice is always ITU.Rec.709 (ITU709 with Sony; SMPTE274 with Panasonic) because it was optimized for HDTV primaries. The matrix does not do a perfect job — that isn't possible — but it can reduce errors to an acceptable minimum. (See Figure 4.)

The chromaticity diagram in Figure 5 shows that all the error vectors are shorter now but that some hue errors have been introduced as part of the trade-off in reducing overall errors.

Many camera users may be familiar with the effects that the matrix can have. Often it is adjusted for effect, but this will always make the colour rendition of the camera less accurate.

Transfer characteristics (gamma)

Strictly speaking, gamma refers only to the display where it describes the relationship between applied signal voltage and light output. For the normal cathode ray tube in a TV set, this is a power law:

L = kVγ

In studio monitors, the value for gamma (γ) is usually around 2.35, but consumer displays generally have a lower value closer to two for economic reasons. LCD and plasma displays are a different matter, but they should emulate this property of the CRT if they are to be acceptable as television displays. (See Figure 6.)

In order to get an overall linear performance, the camera must apply a correction for display gamma, and so it has a gamma-corrector. It is this function in the camera that has the greatest effect on colour performance. Again, HDTV cameras offer several curves to choose from, but the two best curves for colour performance are the ITU.709 curve (#5 in HDW900, #3 in 750/730/730S, “Video Rec” with Panasonic) and the BBC 0.4 curve (#6 in HDW900, #4 in 750/730/730S). (See Figure 7.)

In practice, neither curve fully corrects for the display; a power law of 1/2.35, or about 0.43, has infinite slope near black and would be unusable in a camera since it would grossly amplify electronic noise. The standard curves have a maximum slope near black of 4.5 and five, which is good enough for most purposes.

The chromaticity diagrams in Figure 8 illustrate how important this is. They show the colour performance of an ideal camera (one having perfect colour-matching functions) feeding a monitor with a gamma of 2.35 when trying to reproduce the colours of a GretagMacbeth chart.

Clearly, the BBC curve produces different colour errors, lesser overall. This is mostly because the slope near black is 5X rather than 4.5X as in the ITU.709 curve.

Contrast handling

The standard curves can comfortably cope with a contrast range of about 7.5 stops (180:1), but real scenes contain far more than that. The challenge is to capture more scenic contrast without compromising colour fidelity. Top-end cameras have two helpful features:

  • Black stretchThis expands contrast handling near black by increasing the slope of the curve a little. Typically, the slope can be increased up to about 8X, which effectively adds a stop of contrast handling and improves the accuracy of colour reproduction. Trying to do more will make the picture noisy, but that might be acceptable, depending on the program genre. Figure 9 modifies an ITU.709 curve, raising the slope to eight times; clearly the colour errors are significantly smaller for the saturated colours (those furthest from white).
  • KneeThis is a far more powerful tool. It expands contrast handling near white without affecting darker colours. Camera sensors don't clip at peak white; they can often cope with much more than that. HDTV cameras can easily cope with 600 percent (where 100 percent is normal peak white level) and more in some cameras. Normally, this is an overload condition, prevented by clipping, but it is possible to bend the gamma-corrector curve to capture this range.

A knee point is the point below which normality rules but above which the curve is compressed such that it can capture the 600 percent or so of contrast range that the sensors deliver. This means that highlights such as clouds, speculars and practicals can be included without clipping. This doesn't necessarily improve the colour handling, but it grossly increases the believability of the pictures. The Panasonic Varicam and later Sony HD cameras do this very well, mimicking film performance in many ways. Figure 10 shows the sort of curve that these cameras can supply. Sometimes there are two knees. The less sharp the knee, the better the overall colour rendition.

Interestingly, if these curves are plotted logarithmically, they look rather like the famous “S” curves so admired in film stocks. Not surprisingly, much HDTV drama is shot using a curve like this. (See Figure 11.) This is when the DoP wants a specific film look, such as capturing up to 12 stops of contrast rather than the seven or so of a normal camera.

Alan Roberts consults on HDTV, cameras and colour science.