According to their purpose, convolution operations can be divided into two groups - image smoothing (noise suppression) and gradient operations (edge enhancement).
The equation below describes a two-dimensional discrete convolution with a convolution kernel a. This convolution kernel is sometimes referred to as a convolution mask, and in filter
theory as the impulse response of the filter. The appropriate selection of the convolution kernel is usually based on experiments or knowledge of the frequency properties of the image signal.
|

|
|
Original image
(dimension 320 x 240)
|
|

|
1/18
.
|
20
|
1
|
1
|
1
|
20
|
1
|
-8
|
-8
|
-8
|
1
|
1
|
-8
|
-10
|
-8
|
1
|
1
|
-8
|
-8
|
-8
|
1
|
20
|
1
|
1
|
1
|
20
|
|
|
Image after convolution
Used kernel
|
|
|