Convolution is what is called cross-correlation in signal processing. But is in Convolutional Neural Networks the act of applying a Kernel to an input-image.
Several Convolutions are performed on the input image by moving the Kernel along the pixels. One can pad the image with extra pixels, which could be zeroes or copies of the edge pixels so, that the kernel does not stop moving at the edge of the original image. This will apply the Kernel exactly times.
Example
 A Kernel is applied to an image of the digit 4 with black-white pixels displayed as 0’s and 1’s.
Applying the kernel with matrix multiplication to the top-left pixels yields the value for the corresponding top-left entry in the convolution matrix. This convolution matrix is now a pattern of weights representing the vertical lines of the picture. From the resulting convolution, we see that the Stride is .