Before we can understand what convolution is, it is important to first understand filtering.
Suppose we have a 9 x 9 image as our input, and we need to classify the image as an X or an O. The following diagram illustrates some sample input images.
A perfectly drawn O is shown in the leftmost box in the following diagram, while the other two boxes show badly drawn Os:
A perfectly drawn X is shown in the leftmost box, while the other two boxes show badly drawn Xs:
In either case, we cannot expect the figures to be drawn ...
