Pixotope Keyer
Introduction
Separating actors and props from a green screen scene is a critical function for any virtual-set or mixed-reality project. The aim is to:
identify and isolate the foreground elements from the background screen color
preserve:
fine edge details such as hair
transparency of translucent objects such as glass bottles
remove any spill color from the actors
recover the shadows from the actors and props
The foreground objects need to be completely solid where there is no transparency.
The quality of the produced key is highly dependent on a good keying setup.
Here are some basic rules.
The green screen
The screen should be evenly colored if painted. Ensure a proper chroma green paint is used, and that the surfaces are free from brush marks or other recognisable “texture”
The screen must be stretched tightly to avoid creases if it is made of fabric
Ensure the floor color is the same color as the walls, and that it is clean and without marks
The walls should be at least 3 m high to allow keying of standard-height actors - taller props or actors would require higher walls
The walls should meet the floor seamlessly without creases or shadows. Often, coving is used at the junctions to smooth the transition
The walls should be lit evenly with a wash of light and with no visible highlights or hot spots
The floor should be lit evenly
The subjects should be picked out from behind by a rim light to give them some separation from the screen and to minimise edge spill
The front of the subjects should be illuminated with beauty (key and fill) lights
The camera
Ensure that the camera has a properly calibrated lens and that it is positioned in the desired position on the set
It is normal to have the camera at the same height as the actors' heads so that it is easy to key the actors if they move some distance from the green screen towards the camera
Make sure the lens has been back-focussed.
Make sure proper white balance is set
Make sure the scene is correctly exposed
Ensure the camera is not set to a high gain, which might introduce noise
Check the camera's "clean" video signal on a monitor to ensure there is minimal noise in the image
Do not use image sharpening on the camera if this can be avoided. It may well produce edge artifacts on the key. The image can always be sharpened after the keying process.
NOTE: Noise is an inherent part of any camera video image, and minimizing this is critical for good images as well as creating a good key.
The main issue is to not pick up too much spill color from the environment onto the actors, which would make keying more difficult.
The Talent and Lighting
Do not position them too close to the walls, ideally no closer than 2m and never closer than 1 m
Light them from behind to create good edge definition- rim lights
Light them from the front to create attractive lighting, and to remove any under-illuminated hollows in skin and clothing that could show the screen color - key and fill lights
Position them far enough from the camera to be fully framed with the choice of lens and zoom setting
Talent should not wear any clothing that includes the same color as the screen color
Also be careful of colours that are close to the background color e.g. yellow or cyan for a green screen; cyan or magenta for a blue screen. These may be problematic for de-spill.
Try to avoid transparent clothing that might pick up and reflect the background color
Talent should avoid clothing textures that might cause moiré patterns e.g. herringbone, plaid or dog-tooth patterns
Video keyer
The video keyer is an easy-to-set-up, real-time chroma keyer that you can use to key your video input. It is a component implemented inside the Pixotope Video Pipeline. The keyer algorithm extracts an alpha mask from the received frame, allowing the removal of the specifically picked color from the background of the scene and from its reflections in the foreground image. The user interface for the keyer is included in the Pixotope Director under Production → Adjust → Video Keyer.
Color management
Pixotope’s Video keyer runs in linear ACEScg and can therefore be used on any input color space, such as S-Log, Rec.709, Rec.2020, etc.
Linear ACEScg is not easy to work with on a normal monitor. The image shown when doing annotations is therefore converted into an intermediate space. This image might differ from the final output, but it is only used for viewing the annotations.
Learn more about how to Use the video keyer