In the same vein, you could also do it with variable line heights in digital books, down to pixel differences. At the very base you only need 27 differences to

In the same vein, you could also do it with variable line heights in digital books, down to pixel differences. At the very base you only need 27 differences to identify each characters right? That could probably be reduced to 5 if you combine with other factors.