Syntax :
	Gauss(clip source, float sigma, float power, int rad, int color)
 
Defaults : 
	Gauss(4.0,1.0,ceil(sigma*2),(power>0)?1:0)

Operation :

	Gaussian blur, that is, convolution by a kernel
	G(x,y) = exp(- (x^2+y^2)/(2*sigma^2) )

	The kernel is thresholded by the "rad" argument :
	K(x,y) = G(x,y) if Ninf(x,y) <= rad
	K(x,y) = 0 if Ninf(x,y) > rad

	Where Ninf is the uniform norm, Ninf(x,y) = max(|x|,|y|)
	
	A linear operation is the performed on the convolved result,
	to achieve straigtforward unsharp masking.
	output_pel = input_pel + (input_pel - computed_pel) * power
	
	the color parameter defines on which planes the operation is performed
	color = 0 => performed on Y plane only
	color = 1 => performed on all planes
	color = 2 => performed on UV planes only
	
Exemples :
	- some gaussian blur
	Gauss(2.0)
	- some unsharp masking
	Gauss(1.5,-1.5)
	- Flat kernel blur (aka box blur) (approximation)
	Gauss(1000000.0,rad=2)