Guide To DI
Published : 19th November 2003
...Very interesting paper.
I'm still trying to wrap my mind around some of the topics. One point they seemed to contradict themselves on, or maybe I'm just confused.
In the section on Film Scanning the paper says, "if an Academy mask was used in shooting, only the Academy area should be mapped to the 2048 pixels, not the full frame aperture (resulting in only 1828 pixels per line for the Academy area)."
Then later in the glossary under the definition of Film Formats the paper says "It is generally considered that all scanning is done at full frame size as this avoids the complexity of adjusting the scanner optics or raster size with risks associated with repeatability and stability." In a chart above this statement shows Academy at 1828 which the paragraph above says will happen if you scan full frame for an Academy mask.
To my understanding telecines and scanners are built with the purpose of changing the gate for various film formats, and changing for them is common practice.
Even though the paper says that film negative contains far more resolution and color information than HD, I still sense in the subtext a bias towards HD origination. The paper states that 4K film scanning retains more resolution and color info from the negative than 2K scanning, but that 4K is unnecessary and 2K is more practical.
Then in a paragraph under Shooting HD says "The resolution of an HD frame is also close to that of a 2K film frame." But it would not at all be close compared with a 4K 16bit log scan. Is discouraging 4K scanning lowering the bar in favor of HD origination?
>used in shooting, only the Academy area should be mapped to the >2048 pixels, not the full frame aperture (resulting in only 1828
You will find both in the field. It really depends on the workflow of the facility and how the post pipeline is set up. Most facilities that we work with scan the whole frame and crop afterwards.
>is unnecessary and 2K is more practical.
Most of today's post work is done at 2K. Several tests have shown that the color detail of a release print most of the times is more like 1K than 4K.
Since you need to move 8 times as much data around for 4K frames it is usually not worth the effort. I am not aware of any feature film post pipeline that works at 4K.
4K is more the domain of film restoration where people try to preserve as much of the original frame as possible. A 4K scan can bring back picture details that the restorator needs to reconstruct the frame. These details may be invisible in a 2K scan.
Some facilities also scan at 4K and down sample to 2K immediately afterwards to get a better result.
>is also close to that of a 2K film frame." But it would not at all be close >compared with a 4K 16bit log scan. Is discouraging 4K scanning
First of all, there are HD cameras out there that can output to 10 bit log. So this is indeed pretty close to a 2K frame both in terms of resolution and bit depth. It is not the same as film, especially since the color resolution of current CCD technology relies on interpolation and does not give native full resolution RGB data. But it is close enough for many people to go that route.
16 bit log is uncommon. Usually it's 8 bit lin, 10 bit log (Cineon, DPX), 16 lin or (rarely) floating point (16 bit or 32 bit per channel). The de-facto standard for scans is either Cineons or DPXs at 10 bit log.
Lin Sebastian Kayser IRIDAS
CEO Adalbertstr. 32
D-80799 Munich Germany
There seems to be some confusion as to the relationship between millimetres of film and pixels. The 2048/1828 was defined by Kodak in the early days of digital and should be long gone now. The definition was that 2048 across s35 would correspond to 1828 across academy so instead of squeezing the 2048 into academy the choice was to either cut a part of the image off!?!? Or down-res. to 1828 thus loosing resolution by applying an unnecessary image re-sampling.
If the scanning device is a Spirit (used for most DI's until now) the sensor is 1920 across s35. The best scan is naturally 1920, subsequently a re-record of the same number of pixels across academy for the IN.
Please note that 1920 is aprox. 100.000 pixels more than 1828.