"Typically, kernels are small matrices, and if the desired specific effect of a kernel is known, many computer programmers who work in image processing will know the best numerical values for a kernel. For example a four-by-four identity (here meaning it will not change the image it operates on if used in a convolution) kernel would be [[1,0,0,0],[0,1,0,0],[0,0,1,0],[0,0,0,1]]."