Apparatus and method for generating a three-dimensional representation from a two-dimensional image

1. A method for generating a three-dimensional face representation, comprising:

finding a face image in a two-dimensional image using a two-dimensional face model built from a two-dimensional training set, wherein finding the face image comprises:

annotating the two-dimensional face model with a plurality of model feature annotations;

using a mean observation vector to estimate a face image location;

using a feature search specific to one of the plurality of model feature annotations to estimate a first feature location; and using at least the face image location and the first feature location to find the face image; and

using a three-dimensional face model built from a three-dimensional training set to generate the three-dimensional face representation from the face image.

2. The method of claim 1 , wherein finding the face image comprises iteratively estimating a face image location based on an initial estimate of the face image location and a new estimate of a location of one of a plurality of model feature annotations.

3. The method of claim 1 , wherein finding the face image comprises:

using a feature search specific to another of the plurality of model feature annotations to estimate a second feature location; and

using at least the face image location, the first feature location, and the second feature location to find the face image.

4. The method of claim 1 , comprising building the two-dimensional face model from the two-dimensional training set.

5. The method of claim 4 , wherein building the two-dimensional face model from the two-dimensional training set comprises:

constructing observation vectors from a plurality of manually annotated feature annotations;

normalizing the observation vectors;

constructing a covariance matrix from the normalized observation vectors; and

performing an eigen analysis on the covariance matrix to build the two-dimensional face model.

6. The method of claim 1 wherein using a three-dimensional face model comprises using a three-dimensional face model built from three-dimensional training sets each comprising sets of three modeled coordinate variables.

7. The method of claim 1 , comprising building the three-dimensional face model from three-dimensional training sets each comprising sets of three modeled coordinate variables.

8. The method of claim 1 , further comprising processing a financial transaction in conjunction with the generating a three-dimensional face representation.

9. The method of claim 1 , comprising:

receiving the two-dimensional image at a server from a remote computer; and

transmitting the three-dimensional face representation from the server to the remote computer.

10. A method for generating a three-dimensional face representation, comprising:

building a two-dimensional face model from built from a two-dimensional training set, wherein finding the face image comprises:

annotating the two-dimensional face model with a plurality of model feature annotations;

using a mean observation vector to estimate a face image location;

using a feature search specific to one of the plurality of model feature annotations to estimate a first feature location; and

using at least the face image location and the first feature location to find the face image;

automatically finding a face image in a two-dimensional image using the two-dimensional face model; and

using a three-dimensional face model built from a three-dimensional training set to generate the three-dimensional face representation from the face image.

11. The method of claim 10 , wherein finding the face image comprises:

using a feature search specific to another of the plurality of model feature annotations to estimate a second feature location; and

using at least the face image location, the first feature location, and the second feature location to find the face image.

12. The method of claim 10 , wherein building the two-dimensional face model from the two-dimensional training set comprises;

constructing observation vectors from a plurality of manually annotated feature annotations;

normalizing the observation vectors;

constructing a covariance matrix from the normalized observation vectors; and

performing an eigen analysis on the covariance matrix to build the two-dimensional face model.

13. The method of claim 10 , wherein using a three-dimensional face model comprises using a three-dimensional face model built from three-dimensional training sets each comprising sets of three modeled coordinate variables.

14. The method of claim 13 , comprising building the three-dimensional face model from three-dimensional training sets each comprising sets of three modeled coordinate variables.

15. The method of claim 10 , further comprising processing a financial transaction in conjunction with the generating a three-dimensional face representation.

16. An apparatus for generating a three-dimensional face representation, comprising:

a memory that at least temporarily stores a two-dimensional face model built from a two-dimensional training set and a three dimensional face model built from a three-dimensional training set;

at least one processor that automatically finds a face image in a two-dimensional image using the two-dimensional face model and generates the three-dimensional face representation from the face image using the three-dimensional face model, wherein the at least one processor annotates the two-dimensional face model with a plurality of model feature annotations, estimates a face image location by using at least one mean observation vector, estimates a first feature location by using a feature search specific to one of the plurality of model feature annotations, and finds the face image by using at least the face image location and the first feature location.

17. The apparatus if claim 16 , wherein the at least one processor:

estimates a second feature location by using a feature search specific to another of the plurality of model feature annotations; and

finds the face image by using at least the face image location, the first location, and the second feature location.

18. The apparatus of claim 16 , wherein the one of the additional processor and the at least one processor;

constructs observation vectors from a plurality of manually annotated feature annotations;

normalizes the observation vectors;

constructs a covariance matrix from the normalized observation vectors; and

builds the two-dimensional face model by performing an eigen analysis on the covariance matrix.

19. The apparatus of claim 16 , wherein the three-dimensional model is built from three-dimensional training sets and each three-dimensional training set comprises sets of three modeled coordinate variables.

20. The apparatus of claim 19 , wherein one of an additional processor and the at least one processor builds the three-dimensional face model from three-dimensional training sets and each three-dimensional training set comprises sets of three modeled coordinate variables.

21. The apparatus of claim 16 , wherein said processor processes a financial transaction.

22. A method for generating a three-dimensional face representation, comprising:

building a two-dimensional face model from a two-dimensional training set;

automatically finding a face image in a two-dimensional image using the two-dimensional face model; and

using a three-dimensional face model built from a three-dimensional training set to generate the three-dimensional face representation from the face image;

wherein building the two-dimensional face model from the two-dimensional training set comprises:

(i) constructing observation vectors from a plurality of manually annotated feature annotations;

(ii) normalizing the observation vectors;

(iii) constructing a covariance matrix from the normalized observation vectors; and

(iv) performing an eigen analysis on the covariance matrix to build the two-dimensional face model.

23. A method for generating a three-dimensional face representation, comprising:

building a two-dimensional face model from a two-dimensional training set;

automatically finding a face image in a two-dimensional image using the two dimensional face model; and

using a three-dimensional face model built from a three-dimensional training set to generate the three-dimensional face representation from the face image;

wherein automatically finding the face image in the two-dimensional image using the two-dimensional face model comprises:

(i) annotating the two-dimensional face model with a plurality of model feature annotations;

(ii) using a mean observation vector to estimate a face image location;

(iii) using a feature search specific to one of the plurality of model feature annotations to estimate a first feature location;

(iv) using at least the face image location and the first feature location to find the face image; and

(v) using a three-dimensional face model built from a three-dimensional training set to generate the three-dimensional face representation from the face image.