Java Applet for the Viterbi Algorithm


 This page presents a Java applet illustrating the Viterbi algorithm in text recognition. Just enter a misspelled word (for example: emphssize) and see if the algorithm can retrieve the correct word. The letters that the Viterbi algorithm outputs are enclosed in a blue rectangle.

 You can select the output style either as a trellis representation or as a table showing the 10 letters with the highest a posteriori probability as each letter in the word is analyzed. You can also select different confusion matrices, these confusion matrices were obtained from the literature (Raviv's work) [8]. The transition probabilities (bigrams) were obtained from this site. Since only bigrams for Markov dependencies are used sometimes the algorithm does poorly on correcting misspelled words.

You can see the transition probabilities here. A new browser window will be open.

Also, if you would like to see the confusion matrices while you are using the applet click here.