Hidden markov model (“HMM”)-based user authentication using keystroke dynamics

1. A method comprising:

collecting at a computer keypad, a plurality of first enrollment keystroke vectors from a first enrolling user;

constructing a first Hidden Markov Model (“HMM”) based on the plurality of first enrollment keystroke vectors;

collecting a second plurality of enrollment keystroke vectors from a second enrolling user;

constructing a second HMM based on the second plurality of enrollment keystroke vectors;

collecting a claimed identity from an authenticating user;

collecting a test keystroke vector from the authenticating user;

computing a first probability that the authenticating user is identical to the enrolling user based on the test keystroke vector and the HMM;

computing a second probability that the authenticating user is identical to the second enrolling user based on the test keystroke vector and the second HMM; and

computing a third probability that the test keystroke vector was produced by the claimed identity using an HMM associated with the claimed identity; and performing a user authentication based on a ratio between the third probability and a maximum probability of the first and second probabilities.

2. The method of claim 1 wherein each of the keystroke vectors comprises a plurality of keystroke event timing elements.

3. The method of claim 1 , further comprising: collecting an answer to a supplemental authentication question if the probability is below a predetermined threshold.

4. The method of claim 1 , further comprising: granting restricted access rights if the probability is below a predetermined threshold.

5. The method of claim 1 wherein the plurality of enrollment keystroke vectors and the test keystroke vector are keystroke key press duration vectors, and wherein the keypad is one of a keyboard and a touch-screen.

6. A method comprising:

collecting at a computer keypad, a sequence of observations of a unidirectional stochastic process;

computing parameters of a first Hidden Markov Model (“HMM”) based on the sequence of observations through an algorithm with complexity O(N 2 T), where N is a number of sub-states per physical state, and T is a number of physical states;

computing parameters of a second HMM based on the sequence of observations through the algorithm with complexity O(M 2 T), where M is a number of sub-states per physical state, T is a number of physical states, and M is not equal to N; and

using the first HMM to compute a probability that a test observation sequence would be produced by a system described by the first HMM, wherein the sequence of observations of the unidirectional scholastic process are a plurality of keystroke dynamics key press duration measurements; and

preparing a consolidated HMM based on the parameters of the first HMM and the parameters of the second HMM.

7. The method of claim 6 wherein the HMM is a first HMM, the method further comprising:

computing parameters of a second HMM based on a second sequence of observations; and selecting one of the first HMM or the second HMM as more likely to represent a system that produced the test observation sequence.

8. The method of claim 6 wherein the sequence of observations comprises a plurality of keystroke dynamics measurements.

9. The method of claim 6 wherein the sequence of observations comprises digitized audio samples of a speaker reciting a predetermined phrase.

10. The method of claim 6 wherein the sequence of observations comprises features extracted from a signature.

11. The method of claim 6 wherein the keypad is one of a keyboard and a touch-screen.

12. The method of claim 7 wherein the first HMM represents a first person, the second HMM represents a second person, and selecting one of the first HMM or the second HMM is identifying a user who produced the test observation sequence.

13. A machine-readable non-transitory medium containing data and instructions to cause a programmable processor to perform operations comprising:

collecting enrollment keystroke dynamics timing measurements as a user types a predetermined string;

preparing a Hidden Markov Model (“HMM”) based on the enrollment keystroke dynamics timing measurements, wherein preparing the HMM comprises constructing a three-dimensional state transition matrix, where a first dimension of the matrix corresponds to a plurality of sub-states of a first character of the predetermined string, a second dimension of the matrix corresponds to a second character of the predetermined string, and a third dimension of the matrix corresponds to a plurality of characters of the predetermined string;

collecting authentication keystroke dynamics timing measurements as a candidate types a predetermined string; and using the HMM to estimate a probability that the candidate is identical to the user, wherein the enrollment keystroke dynamics timing measurements and the authentication keystroke dynamics timing measurements comprise key press duration for each character of the predetermined string.

14. The machine-readable non-transitory medium of claim 13 wherein the HMM has a plurality of sub-states corresponding to each character of the predetermined string.

15. The machine-readable non-transitory medium of claim 13 , containing additional data and instructions to cause the programmable processor to perform operations comprising: granting access to the candidate if the probability exceeds a predetermined threshold.

16. The machine-readable non-transitory medium of claim 13 , containing additional data and instructions to cause the programmable processor to perform operations comprising: performing a supplemental authentication process if the probability is below a predetermined threshold.

17. The machine-readable non-transitory medium of claim 13 , wherein the keypad is one of a keyboard and a touch-screen.