I've seen this effect happen a lot in front of green screens. Is this the same thing?
keenan
@small_potato_ In effect, yes: when you grab pixels from video that was captured in front of a green screen, the color of pixels near the silhouette have effectively been averaged with green already. So, even if you have an accurate alpha channel, premultiplying the color value with this alpha channel will yield pixels with a green tint. Worse, you may have "green spill," meaning that light bouncing off the green screen actually lights your subject (see below). This is even harder to get rid of; I suspect there are some good techniques from computational photography (these days, probably learning-based) to tackle these issues in a nice way.
I've seen this effect happen a lot in front of green screens. Is this the same thing?
@small_potato_ In effect, yes: when you grab pixels from video that was captured in front of a green screen, the color of pixels near the silhouette have effectively been averaged with green already. So, even if you have an accurate alpha channel, premultiplying the color value with this alpha channel will yield pixels with a green tint. Worse, you may have "green spill," meaning that light bouncing off the green screen actually lights your subject (see below). This is even harder to get rid of; I suspect there are some good techniques from computational photography (these days, probably learning-based) to tackle these issues in a nice way.