This is especially helpful if we have a strong colour presence/impression on the green screen. White Balance- Another thing we can do to prep the plate is adjust the white balance. We can do this using a simple grade node to alter the shot while viewing the luminance channel. When we remove green, we remove luminance. So we have to compensate for it later. Luminance- One thing to be cautious of during green screen extraction is luminance. Now we add that back over the original plate using a mask. Now that we have an average of both colours, we can subtract it from the original plate to extract the uneven variation from the shot. We then merge these together using a Merge(average) to get a mid value. We start by creating two Constant nodes for the lightest and darkest shades of green in the shot. The first one we looked at was the IBK method. There are a couple ways to do this if the green screen is uneven. IBK- It’s important to have an even and consistent green screen before starting keying process. Keying with Shuffle node Prep for GS Extraction:ĭenoise- The very first thing we need to do is of course denoise the plate. Using the red channel as a base, we can sometimes tweak other maps and Merge our results to get a clean matte. For example, the red channel is usually good for contrast because red is most prevalent in skin. The Shuffle node– That being said, we briefly demonstrated how the Shuffle node can be used to extract a channel if there’s some contrast there that you can key out. The Luma key deals with luminousity, while the chroma key works best for keying out colours like red, green or blue. There are two main ways to key a green screen- Luma key and chroma key. This time we had a more in-depth survey of the processes involved, also covering some new techniques on how to get a working key, besides the usual keyer node.įirst, we discussed a few points and reminders to note about GS Extraction and Prep.Īs we know, extraction by keying works by finding contrast in a shot which can then be manipulated to form a mask that can be used to exclude the undesired colour. Objects don’t move randomly over time, but noise does.In today’s lesson, we revisited the topic of green screen extraction. Video is a changing image, it moves over time. The answer lies on the very nature of video. How can we know that bright spot on the beach is caused by a sand grain or a noisy pixel? What’s different between a video of a sand beach and pure noise? They both have small detail, the pattern is random in most cases and it affects all of the channels. Our goal is to find a criteria which helps us to separate the damage pixels. A median filter is more subtle, but it will only work when the noise is very visible and very small. A blur filter will remove all the noise, but it will also destroy the rest of the image. You could try to look for small pixels which differ from the surrounding patterns, you could try to smooth the hole images keeping the borders intact, you could apply complex statistical models which decide if a pixel is susceptible to be noisy, etc.īut we do not want to affect the detail and textures of the image. There are many ways to decide how much noise a pixel has.
0 Comments
Leave a Reply. |