That's a much more profound explanation than the given. You can actually come up with values yourself then, and it reveals why cases like "edge detection" and "blur" work so nicely (edge detection approximates differentiation; blur acts as a low pass filter;...).