Method for performing image based regression using boosting

1. A method for performing image based regression using boosting to infer a value of an entity that is associated with an object in an image, the method comprising:

learning a regression function using boosting for a plurality of images in which for each image the associated entity is known, wherein the regression function is configured to predict a value of the entity associated with the object from a continuous range of values; and

using the learned regression function to predict the value of the entity associated with the object in an image in which the entity is not known,

wherein learning the regression function using boosting further comprises:

approximating the regression function as an additive function;

at each iteration of the boosting, adding at least one weak function having a plurality of features corresponding in number to the number of dimensions of the object into the regression function; and

retaining the added weak functions that improve the predictability of the entity by the iterated regression function,

wherein the method is performed by a processor.

2. The method of claim 1 , wherein each feature is represented as a Haar filter.

3. The method of claim 1 , wherein the object is multi-dimensional.

4. The method of claim 1 , wherein using the learned regression function further comprises:

evaluating the weak functions in the regression function; and

combining all the weak functions to infer the value of the entity associated with the object.

5. The method of claim 4 , wherein the object is a human face and the inferred entity is age.

6. The method of claim 4 , wherein the object is an anatomical structure and the entity is position and anisotropic spread of a tumor.

7. The method of claim 4 , wherein the object is a human heart and the entity is location of an endocardial wall.

8. The method of claim 1 , wherein the cost function comprises an L 2 loss function and a regularization term.

9. The method of claim 1 , wherein the features are selected using one of a greedy approach, incremental approach, or an independent approach over output dimensions of the object.

10. The method of claim 1 , wherein the continuous range of values includes at least three discrete values.

11. A method for performing image based regression to infer a value of an entity that is associated with an object in an image, the method comprising:

associating a plurality of features of training images to each one of a plurality of weak functions that models a range of values of an entity of an object in an image, the number of features of each weak function corresponding to the number of dimensions of the object;

iteratively using boosting to learn a regression function from the weak functions, wherein the regression function is configured to predict a value of the entity associated with the object in an image having the object where the value of the entity is unknown; and

using the learned regression function to predict the value of the entity of the object in a new image, wherein the method is performed by a processor.

12. The method of claim 11 , wherein using the learned regression function further comprises:

evaluating the weak functions in the regression function; and

combining all the weak functions to infer the value of the entity associated with the object.

13. The method of claim 11 , wherein each feature is represented as a Haar filter.

14. The method of claim 11 , wherein the object is a human face and the entity is age.

15. The method of claim 11 , wherein the object is an anatomical structure and the entity is position and anisotropic spread of a tumor.

16. The method of claim 11 , wherein the object is a human heart and the entity is a location of an endocardial wall.

17. The method of claim 11 , wherein the features are selected using one of a greedy approach, incremental approach, or an independent approach over output dimensions of the object.