Tuesday 29 January 2013

The Expanding Horizon of Digital Cinematography


Recently I have been working with Faculty of Engineering at University of Bristol and BBC Research and Development to shoot some tests for the expanding capabilities of Digital Cinematography. My role on this project was to oversee the cinematography. In essence we have been trying to calibrate Higher Dynamic Range, Higher Frame Rate and Higher Resolutions to match the eye/brain pathway to create highest immersion for the viewing experience. The first tests were shot in November 2012 and since then have been ongoing.

In the human optical system we have a sensitivity of 14 orders of magnitude of which we can always access 5 orders of magnitude. These 5 orders slide upwards for bright desert sun and downwards to cope with low light levels of moon and starlight. The University of Bristol has a display that exhibits this dynamic range.

Digital Cinematography cameras capture around 12 – 14 stops and 35mm film captures around 18 stops. Standard displays, TV’s projectors & computer screens display between 10 and 16 Fstops – that’s between 2 to 3 orders of magnitude of the entire 14 available in the human system, so if you create a higher dynamic range image of 18 or above stops – it will display beneath its dynamic range on contemporary displays. The common response from people seeing this is: “the image looks ‘plastic’”.

The department of Experimental Psychology at Bristol has already undertaken 2D and 3D immersion tests – but these require of 30 minutes of footage. With a limited budget we decided to shoot the Somerset Carnival because of its high internal illumination and floats with internal movement. To shoot HDR an Epic would shoot 50 fps at 4k in HDR mode with a 4 to 6 stop difference giving 18 stops dynamic range.

We then considered mounting 2 Phantoms to shoot 200 fps at 2k in a 3D mirror rig, one camera exposing the high stop the other the low. Dr Marc Price & Alia Skeikh of BBC R&D placed two cameras on a rig at BBC London, but found that artifacts became evident as the tolerances necessary for HDR alignment are far higher than 3D because you need pixel accurate registration to eliminate these artifacts.

We calculated on Epic we could record 6 mins of 4k, 50 fps HDR (thats 100 fps) that would take 40 minutes to download. We decided that due to much higher levels of data output on the Phantom, we would shoot selected floats at the carnival exhibiting high levels of motion. Had we shot Phantom in a mirror rig at its highest speed you could easily generate 1 terabyte of data per minute and that would take 6 and half hours to download.

An HDR stop-motion test conducted by Aaron Fang of University of Bristol Engineering revealed that you need 7 exposures combined to display full higher dynamic range on the display. So using Red’s strategy of setting a correct Fstop to build upon for HDR did not exploit the full-potential available.

The hardest thing to expose in cinematography is a subject that emits light: At our first shoot in Burnham on Sea on the Epic I set the ASA at 320 at 50 fps with a shutter of 100th, then used a spotmeter to calculate a stop. But following this through to display, we discovered that a full HDR image was not achieved using normal cinematographic judgment. I realised that the 100 year old maxim of exposing to protect highlights was no longer a correct rule for HDR. In fact you had to expose the ‘correct’ stop ‘virtually’ – What I mean by this is that if the Fstop should be F5.6 then we would have set the Iris to overexpose three stops over at F2 plus HDRx highlight protection of three stops under: making 18 stops in total. That might seem obvious now, but on the shoot, sphincters tightened, because the Epic images looked terribly overexposed.

We had planned a second shoot at Wells and collected material at 2 stops and 3 stops over and under e-exposed. In HDR terms that’s between 16 to 18 stops respectively. We are about to shoot more footage for immersion testing at the newly built lab at Bristol University. It should be remembered that the first tests worldwide with this expanded digital cinematographic form took place in the South West of the UK.

From the cinematographers viewpoint, it seems to me that counter-intuitively, lower lighting levels are where HDR will function with most impact on a photographic level. Obviously in the highlights colour formation will be held better than in standard systems, but this is a technical issue, rather than specifically an artistic one. But this is interesting because it may in fact require a renaissance of cinematic judgment – someone will need to know that the end result will be fine. With regard low light, the way we read an image may allow the cinematographer to offer clues when underscoring plot, story & emotional cues, in a far more subtle way than with standard dynamic range that the cinematographic arts have used for decades: that of trying to represent 5 orders of magnitude in a 2 – 3 orders of magnitude display.