Conclusions

In this tutorial the Viterbi Algorithm was introduced under the context of Text Recognition. The Viterbi algorithm is a computationally simple method of finding the maximum a posteriori estimate of a sequence of observations. With English text modeled as a Markov process the text recognition problem can be solved using the Viterbi algorithm.

We can conclude that:

The performance of the Viterbi algorithm greatly depends on the likelihood and transition probabilites between English characters, the more contextual information those probabilities could extract the better the performance of the Viterbi algorithm.

All Markov methods can be characterized as requiring little storage and computation but exhibiting mediocre error correction capability. This is in contrast to dictionary look-up methods which correct errors very well but at the price of excessive computation and storage.

I hope this tutorial was interesting. If you have comments, suggestions or if you find mistakes in the tutorial please send me an email to: latorres@cim.mcgill.ca