The skinny on: Bit-depths
One of the terms or jargon that always crop up in the world of digital filmmaking is ‘bit-depth’. Despite its widespread use in describing cameras, editing programmes and even in monitors, it still remains a mystery as to what bit-depth defines, or how it applies to the digital video world.
Breaking down that terminology, we have ‘bit’ â€“ which in a digital sense is a term for measurement. Computers store binary information in 1s and 0s â€“ or bits â€“ and eight of these (01001011) make a byte. In a picture with 1-bit depth, the 1s will represent white and the 0’s will represent black, and in a logical arrangement, these pixels would form an image.
Greyscaled images that require the in-betweens of black and white are slightly more complicated. Instead of having just two colours in one bit, eight bits (because computers perform calculations in base 8), or a string of 1s and 0s, will form the information to represent one shade. With this, it is possible to represent an spectrum of one colour.
Unlike print, which breaks the colour image down into CMYK (Cyan, Magenta, Yellow and Black), video only requires three channels: red, green and blue (RGB). In this vein, a bit-depth of 8 for each channel of R, G and B will yield a total of 16,777,216 colours. Arriving at this number is easy: all you need is to take 2 and put the power of your bit-depth (8, in this case) to it. 28 = 256 â€“ this is the number of colours per channel. Take 256 and multiply it by itself for each channel (256 x 256 x 256) and you’ll get over 16 million colours in just 8-bits.
Most monitors are 8-bits, which is why your computer refers to your display setting as millions of colours, but refer to it as 24-bit true colour. This is technically the bits per pixel, and not per channel. To prevent confusion, the acronym BPP has been adopted. In all other cases, such as in this article, X-bit simply refers to the number of bits per channel.
Most formats that we commonly work with are 8-bit in nature, and this includes DVCPro, HDCAM, AVCHD and HDV. Higher end cameras and decks such as the new AVC-Intra lines, Panasonic’s mastering standard D5 and the HDCAM SR are at a higher 10-bit. The much hyped about Red One captures 4K at a stunning 12-bit, while the super slow motion Phantom (most recently used in upcoming film The Kallang Roar) captures at a stunning 14-bit.
While the increment from 8 to 10 or 12 may seem small, the number of colours increase exponentially from the minute 2-bit jump. At 8-bits we get a paltry 256 colours, but add 2 more bits and we’ll get 1,024 colours per channel; at 12-bits you’ll get a whopping 4,096 and at 14 you’ll have… to figure it out yourself. But you know the formula and you get the point â€“ a simple move from 8 to 10 bit and you’ll get an extra 1,056,964,608 colours to work with. While it is widely believed that the human eye can only discern about 10 million colours, the extra data comes especially in handy during post when you require more colour detail for colour correction, DI, matte keying and CG work.
Looking at the example we have above (courtesy of The Leap Years‘ DP, Henry Chung), the high bit-depth of the Red camera allows the sun-drenched outdoors in the background to be retained, instead of being clipped. It also holds up much better in the colour correction process (Stills taken from the movie Walk Along With You, directed by Venus Lai, shot on the Red One).
Log VS Linear
If by now your eyes have not glazed over, we’ll move on to the types of bit-depths. Log and linear are the two most common types, and still the battle rages on as to which is better. To put it simply, Linear is actually the native way a cameras captures images â€“ the light that enters into the sensor is converted into a voltage (linear), which is then charted to each assigned pixel on the Analogue to Digital Converter (ADC, also linear), which changes the analogue signal of a voltage into a digital value (still linear). To attain log values, some other equation is performed after this sequence to assign more numerical values to the darker shadows to capture more detail. In this vein, many argue that linear is always better, since remaining in linear equates to less conversion loss, and also many post-production programmes work better in linear space as well.
However, the human eye does not perceive colour and dynamic range in a linear form, and neither thus film, thus the Log takes the lead when it comes to video that will eventually lead to a film-out, or when scanning negatives into digital files for editing and conform. In Log, more values are assigned to the darker range than the brighter side of the scale. To help you understand better, I’ve written up a chart to compare the assigned values in Log and in Linear to one F-stop.
So unlike our base-10 numbering system, in binary the numbers increases twice going up and halves as it decreases. In this vein, the numbers will go up this way: 2,3,8,16,32,64… etc. In the above example we take 10-bits as our unit, with Linear calculations lined up on the left. You will notice that midway through, at 5, the value is 32. This means that there are 32 values assigned to the lower range, while the higher range gets the values from 32 up till 1,024. In log, however, all 1,024 values are distributed evenly over the 10 stops (1024 / 10), allowing the darker half of the scale a total of 512 values.
Here you’ll see that I have rounded off the values to the nearest number, ignoring decimal points altogether. In the world of bit-depth, putting your materials into high-end finishing programmes such as Color, Da Vinci or Quantel means you can convert your bit-depths to 32-bit floating point. What this means is that the decimal point can exist, and is free to ‘float’ through any of the given values, eliminating the need to round off values, allowing for virtually no colour shift or loss. In 32-bit floating point, it is possible to convert from linear to log without much damage, and allows graphic elements, colour tuning and clean ups to be done more accurately.
Many of the high-end cameras such as the FW900 and the Varicam have a ‘Log’ capture profile and is more commonly known as ‘Cine Gamma’. This basically allows you to improve on the dynamic range in-camera so the linear process of capturing can be converted to log right at the origin.
Putting a linear source into a log space will also move the black point up from 1 to 95, and the white point down to 685 so there is a lot more headroom and toe room for grading later on. This is also good practice when sending information to TV since legal levels are set to 940 for whites and 65 for blacks.
Putting this to your advantage
There is not hard and fast rules to bit-depths, but there are plenty of examples which we can learn from. Many post houses work offline in 8-bit formats to keep file sizes small, and conform in 10-bit log later down the pipeline for mastering and printing to film. Others who remain in video can choose either to output in 12-bit linear (such as the case is for digital projections), or remain in 8-bit even in mastering since it is highly likely the material will be viewed on 8-bit monitors and screens.
If you are going to offline in an Apple ProRes 10-bit codec, you will need to make sure that your monitor is ‘high-end’ enough to handle all 10-bits, likewise if you are going to conform in 10-bit â€“ always make sure that your monitor can handle it! There are currently no computer monitors that are 10-bit, save for the one developed by HP, and broadcast monitors that support 10-bit are usually rather pricey, such as the ones from Cine-tal and eCinema.
Besides your monitors, also check that the cable to which you connect are capable of supporting the bit depth. VGA cables are mostly 8-bit, while the more complicated component cable and the popular HDMI cable are able to carry 10 or 12-bits per channel. DVI cables are also mostly 8-bits, but there are now dual-linked cables that can carry up to 16-bits.
Though some might say that it is pointless to up a 8-bit source video into a 12-bit linear work space or a 32-bit floating point operation in post, doing so will allow you to reduce the amount of deterioration that the video would suffer when you grade it or chroma key it in a lower bit depth. After you are done working your magic with the video, you still have the option of outputting it in 8-bit without further image quantizing.