Kernel

A kernel in Convolutional Neural Networks is a set of weights that correspond to a spatial pattern in an image. In Convolutional Neural Networks the kernel is applied to the input-image to detect certain structural features.

Example

An example of a kernel $k$ that identifies regions of the image where a vertical line is present can be expressed as

$$k=\left[\begin{array}{ccc}0& 1& 0\\ & & \\ 0& 1& 0\\ & & \\ 0& 1& 0\end{array}\right]$$

where the 1’s correspond to a vertical line.