[FFmpeg-user] "Extracting" Alpha Channel

[Stefano Sabatini]

On date Thursday 2011-07-28 12:00:36 +1000, Dan Flett encoded:
> In television graphics or vision mixing, it's often necessary to "extract"
> the alpha channel from a graphics file and turn it into a separate "Key" or
> "Matte" clip.
> 
> The Key video is a grayscale representation of the alpha channel - black is
> fully transparent and white is fully opaque.
> 
> It is possible to do this the using "lut" filter (I use lutrgb):
> 
> -vf "[in] format=rgba, split [T1], fifo, lutrgb=r=minval:g=minval:b=minval,
> [T2] overlay [out]; [T1] fifo, lutrgb=r=maxval:g=maxval:b=maxval [T2]" 
> 
> This filter first converts the input video to the rgba colorspace.
> 
> It then splits the input video into 2 streams (as per the turorial in the
> libavilfter docs).  The "background" stream is set to a fully black frame by
> setting the r,g,b values to "minval".  The foreground stream is set to a
> fully white frame by setting the r,g,b values to "maxval".  In both streams
> the alpha channel is untouched.
> 
> At the end of the first stream, the "overlay" filter overlays the white
> stream over the black one, using the untouched alpha channel as the mask.
> The result is a grayscale representation of the alpha channel - the actual
> rgb video content is thrown away.

Another interesting possibility (simpler, faster) would be to
implement a decomposer and a composer filter, for splitting RGBA
channels into separate gray video streams and re-composing them.