Because the Gaussian kernel is separable, the convolution is computed as two convolutions with length n vector kernels, to yield a reduction in computation time. The vector is computed as
v(j) = c exp( - 0.5 * ( j / sigma )2 )where the index j is taken to be 0 at the center of the kernel vector, and where c is chosen so that the vector weights sum to 1.0. An additive Bias can also be supplied. The additive Bias is used as the initial value when accumulating the sum of the products (image data*kernel value) for each neighborhood.
The Gaussian blurring operation is discussed in:
Digital Image Processing, Gonzales, R.C., Wintz, P., Addison Wesley, Second Edition, 1987, pp 163--173.
Port: Img In
Source frequency domain image.
Additive bias constant.
Degree of blur.
Port: I Size
Kernel size in I direction.
Port: J Size
Kernel size in J direction.
Port: Img Out
Filtered frequency domain image.