Confidence based vein image recognition and authentication
An indexed hierarchical tree search structure converts each registration sample into an equivalent registration model based on the clustering of its registration item descriptors in the leaf nodes of the hierarchical tree. Query item descriptors from a query sample from someone wanting to be recognized are distributed into the hierarchical tree. A query model is defined based on the clustering of query item descriptors at the leaf nodes, and registration and verification are made based on comparison of the query model and the registration models.
1. A method of searching for a query object within an object class, said method comprising:
(a) accessing a collection of unique training samples of multiple training objects within said object class;
(b) defining a separate training set of training item descriptors from each of said training samples;
(c) extracting the training item descriptors from the separate training sets into a single composite collection of individual training item descriptors;
(d) creating a hierarchical tree from said composite collection of individual training item descriptors according to relations in the individual training item descriptors, said hierarchical tree having a plurality of leaf nodes;
(e) accessing registration sets of registration item descriptors, each registration set being defined from a respective registration sample obtained from a registration objects to be registered, said registration object being of said object class, individually distributing each registration item descriptor from each registration set into said hierarchical tree according to said relations defined in the creation of said hierarchical tree, indexing the registration item descriptors clustered within each leaf node to their corresponding registration samples;
(f) defining a separate registration model for each registration sample based on the clustering of its corresponding registration item descriptors in each leaf node;
(g) accessing a query sample from said query object, defining a query set of query item descriptors from said query sample, individually distributing each query item descriptor into said hierarchical tree according to said relations defined in the creation of said hierarchical tree;
(h) defining a query model for said query sample based on the clustering of said query item descriptors in each leaf node; and
(i) using the query model, x, and the registration models to identify as a potential candidate match the registration object, i, that renders the highest confidence i* of matching the query object, defined as i*=arg max i P(x|i)P(i).
2. The method of claim 1 , wherein in step (i), the registration object whose corresponding register model is deemed to most closely match the query model is identified as said potential candidate match.
3. The method of claim 1 , wherein in step (d), said hierarchical tree is a binary tree.
4. The method of claim 1 , wherein the collection of unique training samples are taken from a general population of the object class, and the registration samples are not included in the collection of unique training samples.
5. The method of claim 1 , wherein in step (e), said registration samples are said training sample, and registration item descriptors are different from said training item descriptors.
6. The method of claim 1 , wherein in step (e), said registration item descriptors are different from said training item descriptors, said registration samples are different from said training samples and said registration objects are different from said training objects.
7. The method of claim 1 , wherein in step (e), multiple registration samples are obtained from the same registration object in a many-to-one manner.
8. The method of claim 1 , wherein in step (f), each registration model uniquely identifies its corresponding registration sample.
9. The method of claim 1 , wherein in step (f), each registration model is a histogram the numbers of its corresponding registration item descriptors in each leaf node.
10. The method of claim 1 , wherein in step (h), said query model is defined in a manner analogous to how said registration models are defined.
11. The method of claim 1 , wherein in step (i),
IF a measure of the similarity between the query model and a Gaussian mixture of the registration models that correspond to the potential candidate match is greater than a first threshold, THEN said query object is deemed to be registered and correspond to any one of the registration objects.
12. The method of claim 11 , wherein in step (i),
IF said measure of the similarity is further greater than a second threshold higher than said first threshold, THEN said query object is deemed to be authenticated and specifically correspond to the potential candidate match.
13. The method of claim 11 , wherein in step (i),
IF said measure of the similarity is not greater than said first threshold, then said query object is rejected as not corresponding any registration object.
14. The method of claim 1 , wherein,
IF i* is greater than a first threshold, THEN said query object is deemed to be registered and correspond to any one of the registration objects.
15. The method of claim 1 , implemented in a registration verification system having at least one data processing unit configured to execute steps (a) through (i).
16. The method of claim 1 , further including destroying said training samples and said training sets of training item descriptors.
17. The method of claim 1 , wherein said object class is a human person, and said training samples, registration samples and query samples are biometric samples of a human person.
18. A method of searching for a query object within an object class, said method comprising:
(a) accessing a collection of unique training samples of multiple training objects within said object class;
(b) defining a separate training set of training item descriptors from each of said training samples;
(c) creating a composite collection of training item descriptors from the separate training sets of sample item descriptors;
(d) creating a hierarchical tree from said composite collection of training item descriptors according to relations in the training item descriptors, said hierarchical tree having a plurality of leaf nodes;
(e) accessing registration sets of registration item descriptors defined from respective registration samples obtained from registration objects of said object class, distributing said registration sets of registration item descriptors into said hierarchical tree according to said relations defined in the creation of said hierarchical tree, indexing the registration item descriptors clustered within each leaf node to their corresponding registration samples;
(f) defining a separate registration model for each registration sample based on the clustering of its corresponding registration item descriptors in each leaf node;
(g) accessing a query sample from said query object, defining a query set of query item descriptors from said query sample, distributing said query set of query item descriptors into said hierarchical tree according to said relations defined in the creation of said hierarchical tree;
(h) defining a query model for said query sample based on the clustering of said query item descriptors in each leaf node; and
(i) using the query model, x, and the registration models to identify as a potential candidate match the registration object, i, that renders the highest confidence i* of matching the query object, defined as i*=arg max i P(x|i)P(i);
wherein in step (i):
each registration object i corresponds to M i registration models;
said query model and said registration models are in matrix form;
the registration models are denoted μ ij where i specifies a registration model's corresponding registration object and j identifies a specific one of the multiple registration models 1−M i that correspond to registration object i;
the registration samples that corresponds to the same registration object are described by a Gaussian mixture model, as follows:
P ( x|i )˜{ w ij ,N ( x|μ ij ,σ ij 2 )}∀ j= 1 , . . . M i
where N is the normal distribution, σ is the variance, and w ij =1/M i ; and
the registration object i whose corresponding confidence i* satisfies the relationship i*=arg max i =(max j (xμ ij T )) is deemed said potential candidate match.
19. A method of searching for a query object within an object class, said method comprising:
(a) accessing a collection of unique training samples of multiple training objects within said object class;
(b) defining a separate training set of training item descriptors from each of said training samples;
(c) creating a composite collection of training item descriptors from the separate training sets of sample item descriptors;
(d) creating a hierarchical tree from said composite collection of training item descriptors according to relations in the training item descriptors, said hierarchical tree having a plurality of leaf nodes;
(e) accessing registration sets of registration item descriptors defined from respective registration samples obtained from registration objects of said object class, distributing said registration sets of registration item descriptors into said hierarchical tree according to said relations defined in the creation of said hierarchical tree, indexing the registration item descriptors clustered within each leaf node to their corresponding registration samples;
(f) defining a separate registration model for each registration sample based on the clustering of its corresponding registration item descriptors in each leaf node;
(g) accessing a query sample from said query object, defining a query set of query item descriptors from said query sample, distributing said query set of query item descriptors into said hierarchical tree according to said relations defined in the creation of said hierarchical tree;
(h) defining a query model for said query sample based on the clustering of said query item descriptors in each leaf node; and
(i) using the query model, x, and the registration models to identify as a potential candidate match the registration object, i, that renders the highest confidence i* of matching the query object, defined as i*=arg max i P(x|i)P(i);
wherein for each registration object, a Gaussian mixture is defined for all its corresponding registration models, and an authentication confidence s is defined as
s =max i (max j ( xμ ij T ))
where μ ij denotes the registration models, i specifies a registration model's corresponding registration object, j identifies a specific one of the multiple registration models that correspond to registration object i, and T is the transform operator; and
IF s is greater than a predefined threshold of confidence greater than i*, THEN said query object is deemed to be authenticated and specifically correspond to the potential candidate match.
20. The method of claim 19 , wherein an optimal value of T is defined as
T
*
=
min
T
∑
i
∑
j
=
1
M
i
sign
(
s
ij
-
T
)
-
l
ij
2
where l is the ground-truth label in [−1,1].