Convolutional Filter
Refers to the [see page 14, weights] used for weighting in a moving-average calculation.
Filters can be called non-zero over its region of support to mean that the filter is 0 everywhere outside of the window its being used in.
Filters are often normalised so that it sums to 1.0 (avoids a filter producing a completely different color than was put input to it).
Note: most filters are symmetric.