Take the 2-minute tour ×
Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It's 100% free, no registration required.

If we are to classify 2 separate classes/actions using HMMs, we design 2 separate HMMs (one for each class). Do they share same set or a different set of observations-symbols for each of the HMM?

If they share different set of symbols, during the time of extracting a symbol (using vector quantization) how do we know which set of symbols to generate in order to classify them?

share|improve this question
Can you provide some more information about exactly what your data is and how you plan to use HMMs to classify? –  Nick Jan 21 '12 at 18:03
I'm dealing with action recognition where I have to recognize Action A from Action B using HMMs. I have series of observations extracted from feature vectors(mesh features in a binary image)using k-means clustering. Are the observations symbols common for both the Actions or not is my question? Here is the paper that I'm referring to –  garak Jan 22 '12 at 13:13
You don't want a separate HMM for each class. There is only one HMM for all the classes. –  Aaron Jul 6 at 13:19

1 Answer 1

my advice is to use the same set. if you use 2 different set you have to quantize (= do labeling) the gesture twice (one time for each hmm and use for each hmm the appropriate labeling).

if you use kmeans take care that standard implementation uses random initialization each time it is called (so each time different set of symbol). so, for example in opencv kmeans function, you need do call function with a predefined centroids and o only 1 iteration (so for each point in you gesture-space is labelled with the euclidean distance in the same label-space)

tell me if you need more explanation, hope it helps

share|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.