Method and apparatus for object recognition using probability models
A method and an apparatus automatically recognize or verify objects in a digital image using probability models. According to a first aspect, a method and apparatus automatically recognize or verify objects in a digital image by: accessing digital image data including an object of interest therein; detecting an object of interest in the image; normalizing the object to generate a normalized object representation; extracting a plurality of features from the normalized object representation; and applying each feature to a previously-determined additive probability model to determine the likelihood that the object of interest belongs to an existing class. In one embodiment, the previously-determined additive probability model is an Additive Gaussian Model.
1. A method for automatically recognizing or verifying objects in a digital image, said method comprising:
accessing digital image data containing an object of interest therein;
using at least one processor for detecting an object of interest in said digital image data of interest in said digital image data;
normalizing said object of interest to generate a normalized object representation;
extracting a plurality of features from said normalized object representation; and
applying each extracted feature to a previously-determined additive probability model to determine the likelihood that the object of interest belongs to an existing class of objects,
wherein said additive probability model models the objects using a class center and residual components between the objects and the class center, wherein an uncertainty related to the class center is represented by a model associated with the class center,
wherein said object of interest is a face and said step of detecting the object of interest detects facial features in said digital image data.
2. The method according to claim 1 , wherein said previously-determined additive probability model is an Additive Gaussian Model that decomposes an appearance of an object into components corresponding to class and view.
3. The method according to claim 1 , further comprising:
selecting the existing class for said object of interest based on said likelihood; and
re-calculating the additive probability model for the selected class using a feature value of the object of interest.
4. The method according to claim 1 , wherein said step of detecting the object of interest utilizes early rejection to determine that an image region does not correspond to a facial feature.
5. The method according to claim 1 , wherein said object of interest is a face in a digital photo.
6. The method according to claim 1 , further comprising:
generating an additive probability model for each of a plurality of classes based on feature values for objects belonging to said classes.
7. The method according to claim 6 , wherein said step of generating the additive probability model for a particular class is repeated each time a detected object of interest is added to the corresponding class.
8. The method according to claim 6 , wherein said step of generating the additive probability model clusters examples belonging to a single class so as to generate multiple additive probability models for each class identity.
9. The method according to claim 6 , wherein said step of generating the additive probability model computes a posterior distribution for a feature value mean from at least one example feature value.
10. The method according to claim 9 , wherein said additive probability model models variance of said feature value mean.
11. The method according to claim 10 , wherein said variance of said feature value mean approaches zero as more examples are associated with the corresponding class.
12. The method according to claim 1 , further comprising:
executing a training stage to identify a set of independent features that discriminate between classes.
13. The method according to claim 1 , wherein said digital image data represents a digital photo.
14. An apparatus for automatically recognizing or verifying objects in a digital image, said apparatus comprising:
a digital image data input for accessing digital image data containing an object of interest therein;
an object detector for detecting an object of interest in said digital image data;
a normalizing unit for normalizing said object of interest to generate a normalized object representation;
a feature extracting unit for extracting a plurality of features from said normalized object representation; and
a likelihood determining unit for applying each extracted feature to a previously-determined additive probability model to determine the likelihood that the object of interest belongs to an existing class of objects,
wherein said additive probability model models the objects using a class center and residual components between the objects and the class center, wherein an uncertainty related to the class center is represented by a model associated with the class center, and
wherein said object of interest is a face and said object detector for detecting an object of interest detects facial features in said digital image data.
15. The apparatus according to claim 14 , wherein said previously-determined additive probability model is an Additive Gaussian Model that decomposes an appearance of an object into components corresponding to class and view.
16. The apparatus according to claim 14 , wherein said likelihood determining unit selects the existing class for said object of interest based on said likelihood; and re-calculates the additive probability model for the selected class using a feature value of the classified object of interest.
17. The apparatus according to claim 14 , wherein said object detector detects the object of interest utilizing early rejection to determine that an image region does not correspond to a facial feature.
18. The apparatus according to claim 14 , wherein said object of interest is a face in a digital photo.
19. The apparatus according to claim 14 , wherein said apparatus generates an additive probability model for each of a plurality of classes based on feature values for objects belonging to said classes.
20. The apparatus according to claim 19 , wherein said apparatus repeats generating the additive probability model for a particular class each time a detected object of interest is added to the corresponding class.
21. The apparatus according to claim 19 , wherein said apparatus generates the additive probability model by clustering examples belonging to a single class so as to generate multiple additive probability models for each class identity.
22. The apparatus according to claim 19 , wherein said apparatus generates the additive probability model by computing a posterior distribution for a feature value mean from at least one example feature value.
23. The apparatus according to claim 22 , wherein said additive probability model models variance of said feature value mean.
24. The apparatus according to claim 23 , wherein said variance of said feature value mean approaches zero as more examples are associated with the corresponding class.
25. The apparatus according to claim 14 , wherein said apparatus executes a training stage to identify a set of independent features that discriminate between classes.
26. The apparatus according to claim 14 , wherein said digital image data represents a digital photo.