GeneralConvolution

AviSynth+
Up-to-date documentation: https://avisynthplus.readthedocs.io

Performs a matrix convolution on an RGB32 (AVS+ any format) clip.

Syntax and Parameters

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

AVS+ GeneralConvolution (clip clip, [ float bias, string matrix, float divisor, bool auto, bool luma, bool chroma, bool alpha ] )

clip  clip =

Source clip. Must be RGB32. AVS+ All color spaces since Avisynth+ r2768.

int  bias = 0
AVS+ float  bias = 0.0

Additive bias to adjust the total output intensity.

string  matrix = "0 0 0 0 1 0 0 0 0"

A 3×3 or 5×5 matrix with 3² (9) or 5² (25) integer values.

AVS+ A 3×3, 5×5, 7×7 or 9×9 matrix with 3² (9), 5² (25), 7² (49), 9² (81) float or integer values. float values are converted to integers for 8-16 bit clips

float  divisor = 1.0

Divides the output of the convolution before adding bias.

bool  auto = true

Enables auto scaling. Auto scaling divides the output of the convolution 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 scaling is disabled.

AVS+

bool  luma = true

bool  chroma = true

bool  alpha = true

Enables processing only selected planes. For RGB clips luma and chroma setting is ignored. Unprocessed planes are simply copied. E.g. alpha=false can speed up RGB32 processing, usually A channel is not used.

The divisor is usually the sum of the elements of the matrix. But when the sum is zero, you can leave divisor=1 and use 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 bias, the pixels are clipped to the range 0-255 AVS+ or the appropriate min-max range of the specific bit depth. In 32 bit float formats no clamp happens.

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

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(0, "
    1  2  1 
    0  0  0 
   -1 -2 -1 ", 8)

 

• Vertical (Sobel) Edge Detection:

GeneralConvolution(0, "
   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 to the 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 ", 0, 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) )/1.0 + 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

 

• Emboss filter (3D relief effect)

GeneralConvolution(128, "
-1 0 0
 0 0 0
 0 0 1")

External Links

• AviSynth’s GeneralConvolution explained
• Some other examples can be found here.

Changelog

AVS+ r2768	Supports all color spaces, added 7x7 and 9x9 matrix sizes, added luma, chroma, alpha
v2.55	Added divisor, auto.
v2	Initial Release