Online training for object recognition system

1. A method of online training of a classifier, the method comprising:

determining a distance from at least one feature vector of an object, which is observed during the online training, to a first decision boundary established during off-line training for the classifier, the first decision boundary defined between a true object vector and a false object vector;

updating a decision rule as a function of at least the distance; and

classifying a future example based at least in part on a second decision boundary established from the updated decision rule.

2. The method of claim 1 , in which the function is further based at least in part on a number of observed positive examples, and/or an amount of observed time since a start of the online training.

3. The method of claim 1 , in which the object is a fingerprint, an iris, and/or a face.

4. The method of claim 1 , further comprising reducing a margin between the at least one feature vector of the object and the second decision boundary based at least in part on a number of positive examples that are observed.

5. The method of claim 1 , in which the decision rule is based at least in part on a measured mean and/or a variance of the distance.

6. The method of claim 1 , further comprising iteratively updating the decision rule based at least in part on new data.

7. The method of claim 6 , further comprising stopping updates to the decision rule based at least in part on a first set of conditions.

8. The method of claim 7 , in which the first set of conditions comprise at least a number of samples, and/or an amount of time.

9. The method of claim 7 , further comprising starting updates to the decision rule based at least in part on a second set of conditions.

10. The method of claim 9 , in which the second set of conditions comprise a false rejection rate.

11. The method of claim 9 , in which starting updates to the decision rule comprises resetting, re-using, or backing up at least one learned value.

12. The method of claim 9 , in which starting updates to the decision rule comprises modifying the first set of conditions.

13. The method of claim 1 , further comprising computing and tracking a false rejection rate (FRR) and further updating the decision rule based at least in part on the FRR.

14. The method of claim 1 , in which the function further comprises tracking a plurality of distances and the decision rule is based at least in part on statistics of the plurality of distances.

15. The method of claim 1 , in which the second decision boundary is defined between the first decision boundary and the true object vector.

16. An apparatus for online training of a classifier, the apparatus comprising:

a memory unit; and

at least one processor coupled to the memory unit; the at least one processor configured:

to determine a distance from at least one feature vector of an object, which is observed during the online training, to a first decision boundary established during off-line training for the classifier, the first decision boundary defined between a true object vector and a false object vector;

to update a decision rule as a function of at least the distance; and

to classify a future example based at least in part on a second decision boundary established from the updated decision rule.

17. The apparatus of claim 16 , in which the function is further based at least in part on a number of observed positive examples, and/or an amount of observed time since a start of the online training.

18. The apparatus of claim 16 , in which the object is a fingerprint, an iris, and/or a face.

19. The apparatus of claim 16 , in which the at least one processor is further configured to reduce a margin between the at least one feature vector of the object and the second decision boundary based at least in part on a number of positive examples that are observed.

20. The apparatus of claim 16 , in which the decision rule is based at least in part on a measured mean and/or a variance of the distance.

21. The apparatus of claim 16 , in which the at least one processor is further configured to iteratively update the decision rule based at least in part on new data.

22. The apparatus of claim 21 , in which the at least one processor is further configured to stop updates to the decision rule based at least in part on a first set of conditions.

23. The apparatus of claim 22 , in which the first set of conditions comprise a number of samples, and/or an amount of time.

24. The apparatus of claim 22 , in which the at least one processor is further configured to start updates to the decision rule based at least in part on a second set of conditions.

25. The apparatus of claim 24 , in which the second set of conditions comprise a false rejection rate.

26. The apparatus of claim 24 , in which the at least one processor is further configured to start updates to the decision rule by resetting, re-using, or backing up at least one learned value.

27. The apparatus of claim 24 , in which the at least one processor is further configured to start updates to the decision rule by modifying the first set of conditions.

28. The apparatus of claim 16 , in which the at least one processor is further configured to compute and track a false rejection rate (FRR) and further update the decision rule based at least in part on the FRR.

29. The apparatus of claim 16 , in which the at least one processor is further configured to track a plurality of distances, and the decision rule is based at least in part on statistics of the plurality of distances.

30. The apparatus of claim 16 , in which the second decision boundary is defined between the first decision boundary and the true object vector.

31. A non-transitory computer-readable medium having program code recorded thereon, the program code being executed by a processor and comprising:

program code to determine a distance from at least one feature vector of an object, which is observed during the online training, to a first decision boundary established during off-line training for a classifier, the first decision boundary defined between a true object vector and a false object vector;

program code update a decision rule as a function of at least the distance; and

program code classify a future example based at least in part on a second decision boundary established from the updated decision rule.

32. An apparatus for online training of a classifier, the method comprising:

means for determining a distance from at least one feature vector of an object, which is observed during the online training, to a first decision boundary established during off-line training for the classifier, the first decision boundary defined between a true object vector and a false object vector;

means for updating a decision rule as a function of at least the distance; and

means for classifying a future example based at least in part on a second decision boundary established from the updated decision rule.