GeneralConvolution

GeneralConvolution (clip, int "bias", string "matrix", float "divisor", bool "auto")

This filter performs a matrix convolution.

clipRGB32 clip
bias (default 0)additive bias to adjust the total output intensity
matrix (default "0 0 0 0 1 0 0 0 0")can be a 3x3 or 5x5 matrix with 9 or 25 integer numbers between -256 and 256
divisor (default 1.0)divides the output of the convolution (calculated before adding bias)
auto (default true)Enables the auto scaling functionality. This divides the result by the sum of the elements of the matrix. The value of divisor is applied in addition to this auto scaling factor. If the sum of elements is zero, auto is disabled

The divisor is usually the sum of the elements of the matrix. But when the sum is zero, you must use divisor and the bias setting to correct the pixel values. The bias could be useful if the pixel values are negative due to the convolution. After adding a bias, the pixels are just clipped to zero (and 255 if they are larger than 255).

Around the borders the edge pixels are simply repeated to service the matrix.

Some examples:

# Blur:

GeneralConvolution(0, "
   10 10 10 10 10
   10 10 10 10 10
   10 10 16 10 10
   10 10 10 10 10
   10 10 10 10 10 ", 256, False)
# Horizontal (Sobel) edge detection:

GeneralConvolution(128, "
    1  2  1
    0  0  0
   -1 -2 -1 ", 8)
# Vertical (Sobel) Edge Detection:

GeneralConvolution(128, "
   1  0 -1
   2  0 -2
   1  0 -1 ", 8)
# Displacement (simply move the position
# of the "1" for left, right, up, down)

GeneralConvolution(0,"
   0 1 0
   0 0 0
   0 0 0 ")
# Displacement by half pixel up (auto scaling):

GeneralConvolution(0,"
   0 1 0
   0 1 0
   0 0 0 ")
# Displacement by half pixel right (manual scaling):

GeneralConvolution(0,"
   0   0   0
   0 128 128
   0   0   0 ", 256, False)
# Sharpness filter:

GeneralConvolution(0,"
   0   -1   0
  -1    5  -1
   0   -1   0 ", 1, True)

In this case, the new pixel values y(m,n) are given by
y(m,n) = (-1*x(m-1,n) - 1*x(m,n-1) + 5*x(m,n) - 1*x(m,n+1)
         - 1*x(m+1,n))/(-1-1+5-1-1)/1.0 + 0
# Slight blur filter with black level clipping and 25% brightening:

GeneralConvolution(-16,"
   0   12   0
  12  256  12
   0   12   0 ", 0.75 ,True)

In this case, the new pixel values y(m,n) are given by
y(m,n) = ( 12*x(m-1,n) + 12*x(m,n-1) + 256*x(m,n) + 12*x(m,n+1)
         + 12*x(m+1,n) )/(12+12+256+12+12)/0.75 - 16

Some other examples can be found here.

Changelog:

v2Initial Release
v2.55added divisor, auto

$Date: 2008/05/28 21:24:49 $