I shot this against an 11' x 15' green-screen using two photo floods with remote flashes to backflood the screen, then keyed it over a commercial backdrop from Owen's Originals using WinImages.

The green-screen had some folds and other brightness variations, so I used medium-resolution color space keying, which is an amazingly powerful keying mechanism. Three simple polygonal area selections later — about ten seconds work — and I had a perfect key.

These techniques work exactly the same for a blue-screen, I just prefer a green-screen because I can see it better.

The backdrop was white stone, so I sampled the skin tone and tinted the backdrop so it would carry the same warm tones her skin does. Lastly, I wrote a quick green de-fringe script and ran it against the edges of the keyed subject where the skin tones showed some green due to insufficient back-lighting. The mechanism I chose for pixel(R,G,B) was...

    if G > (R+B)/2 then G = (R+B)/2

...which lets the red and blue channels set additional tonality, if any, while locking input from the green channel out, but only where the green channel was providing the majority of the color in the first place. Technically speaking, this will work perfectly for greyscale subject matter, reds, blues and purples. Other colors will fare less well.

Been thinking about a sample-based de-fringe; this would let you take a sample (eg of skin tone) and this would get you relative G for a particular value of (R and B), and then when you found G > (that relative value) you could insert that relative value. This would track the luma of the tone to some degree, where the current method does not.

This is a bit complex for a script because scripts are slow when processing large masked regions; so I'd probably write it as a plug-in.

Canon EOS 40D, Canon EF-S 18-55 mm ƒ/3.5-5.6 IS zoom @ 55 mm, ƒ/5.6, 1/13th sec exposure, ISO 3200, 2s delayed shutter to stabilize, tripod.

#portrait #winimages

No EXIF Data