According to your explanation, the algorithm does not take into account the shape of each character at all, which is something that could be exploited for example to be able to make out the 8 in the first image, rather than downgrading it to something that looks like a U.