Ken Turkowski Adobe Systems
We present an incremental method for computing the Gaussian at a sequence of regularly spaced points, with the cost of one vector multiplication per point. This technique can be used to implement image blurring by generating the Gaussian coefficients on the fly, avoiding an extra texture lookup into a table of precomputed coefficients.
Filtering is a common operation performed on images and other kinds of data in order to smooth results or attenuate noise. Usually the filtering is linear and can be expressed as a convolution: