Method and apparatus for performing object detection

• US 20070140550A1
• Filed: 12/20/2005
• Published: 06/21/2007
• Est. Priority Date: 12/20/2005
• Status: Active Grant

• Current Assignee: Google Technology Holdings LLC (Alphabet Inc.)
• Original Assignee: General Instrument Corporation (CommScope Holding Company, Inc.)
• Inventors: Li, Zhu, Li, Dongge, Gandhi, Bhavan
• Granted Patent: US 7,697,752 B2
• Time in Patent Office: Days
• US Class Current: 382/159
• CPC Class Codes: G06V 40/165 using facial parts and geometric relationships

Abstract

An object detection algorithm that generates a two-layer Gaussian Mixture Model (GMM) during a training session, and subsequent to the training session, uses the two-layer GMM to perform face detection. No labeling of local features is needed. The only input that is provided by a user is the setting of a few global parameters for the image being captured during the training session, such as, for example, the person'"'"'s facial pose.

Claims

1. An apparatus for generating a model that can be used to perform object detection, the apparatus comprising:

an input interface configured to receive training images; and

a processor in communication with the input interface for receiving traning images from the input interface, the processor being configured to perform a training algorithm, the training algorithm processing multiple training images received from the input interface to produce a Gaussian Mixture Model (GMM) comprising at least a first-layer GMM, the first-layer GMM describing spatial attributes of local features contained in an image, the first-layer GMM receiving local feature detection results from a local feature detection algorithm.

2. The apparatus of claim 1, wherein the local feature detection algorithm is a second-layer GMM describing intensity variation attributes of local features contained in the image.

3. The apparatus of claim 2, wherein the first-layer GMM is generated by using one or more global parameters associated with the training images to calculate Gaussian spatial distributions for the training images, the first logic combining the Gaussian spatial distributions to produce the first-layer GMM.

4. The apparatus of claim 3, wherein said one or more global parameters include a pose.

5. The apparatus of claim 3, wherein said one or more global parameters include a size of the training image.

6. The apparatus of claim 3, wherein said one or more global parameters include a distance of the training image away from an image capture device that was used to capture the training image.

7. The apparatus of claim 2, wherein the second-layer GMM is generated by using pixel gray scale intensity values of regions of the training image that include local features to calculate Gaussian gray scale distributions for the local features, the first logic combining the Gaussian gray scale distributions to produce the first-layer GMM.

8. An apparatus for performing object detection, the apparatus comprising:

an input interface configured to receive images; and

a processor in communication with the input interface, the processor being configured to perform a face detection algorithm, the algorithm processing an input image received from the input interface to determine a probability that the input image fits a Gaussian Mixture Model (GMM) comprising at least a first-layer GMM that describes spatial attributes of local features contained in an image, and a local feature detection algorithm, the first-layer GMM receiving local feature detection results from the local feature detection algorithm

9. The apparatus of claim 8, wherein the local feature detection algorithm is a second-layer GMM that describes intensity variation attributes of local features contained in the image.

10. The apparatus of claim 9, wherein the first-layer GMM is generated by using one or more global parameters associated with training images to calculate Gaussian spatial distributions for the training images, the first logic combining the Gaussian spatial distributions to produce the first-layer GMM.

11. The apparatus of claim 10, wherein said one or more global parameters include a pose.

12. The apparatus of claim 10, wherein said one or more global parameters include a size of the training image.

13. The apparatus of claim 10, wherein said one or more global parameters include a distance of the training image away from an image capture device that was used to capture the training image.

14. The apparatus of claim 10, wherein the second-layer GMM is generated by using pixel gray scale intensity values of regions of the training image that include local features to calculate Gaussian gray scale distributions for the local features, the first logic combining the Gaussian gray scale distributions to produce the first-layer GMM.

15. A method for training a model to perform object detection comprising:

processing multiple images in accordance with a training algorithm to produce a multi-layer Gaussian Mixture Model (GMM) comprising at least first layer GMM and a second-layer GMM, the first-layer GMM describing spatial attributes of local features contained in an image, the second-layer GMM describing intensity variation attributes of local features contained in the image.

16. The method of claim 15, wherein the first-layer GMM is generated by using one or more global parameters associated with the training images to calculate Gaussian spatial distributions for the training images, and by combining the Gaussian spatial distributions to produce the first-layer GMM.

17. The method of claim 15, wherein said one or more global parameters include a pose.

18. The method of claim 15, wherein said one or more global parameters include a size of the training image.

19. The method of claim 15, wherein said one or more global parameters include a distance of the training image away from an image capture device that was used to capture the training image.

20. The method of claim 15, wherein the second-layer GMM is generated by using pixel gray scale intensity values of regions of the training image that include local features to calculate Gaussian gray scale distributions for the local features, the first logic combining the Gaussian gray scale distributions to produce the first-layer GMM.

21. A method for performing object detection comprising:

processing an input image to determine a probability that the input image fits a multi-layer Gaussian Mixture Model (GMM) comprising at least first layer GMM and a second-layer GMM, the first-layer GMM describing spatial attributes of local features contained in an image, the second-layer GMM describing intensity variation attributes of local features contained in the image.

22. The method of claim 21, wherein the first-layer GMM is generated by using one or more global parameters associated with training images to calculate Gaussian spatial distributions for the training images, and by combining the Gaussian spatial distributions to produce the first-layer GMM.

23. The method of claim 21, wherein said one or more global parameters include a pose.

24. The method of claim 21, wherein said one or more global parameters include a size of the training image.

25. The method of claim 21, wherein said one or more global parameters include a distance of the training image away from an image capture device that was used to capture the training image.

26. The method of claim 21, wherein the second-layer GMM is generated by using pixel gray scale intensity values of regions of the training image that include local features to calculate Gaussian gray scale distributions for the local features, and by combining the Gaussian gray scale distributions to produce the first-layer GMM.

27. A computer program for generating a model that can be used to perform object detection, the computer program being embodied on a computer-readable medium, the program comprising:

code for performing a training algorithm, the training algorithm processing multiple training images to produce a multi-layer Gaussian Mixture Model (GMM) comprising at least first layer GMM and a second-layer GMM, the first-layer GMM describing spatial attributes of local features contained in an image, the second-layer GMM describing intensity variation attributes of local features contained in the image.

28. A computer program for performing object detection, the program being embodied on a computer-readable medium, the program comprising:

code for processing an input image to determine a probability that the input image fits a multi-layer Gaussian Mixture Model (GMM) comprising at least first layer GMM and a second-layer GMM, the first-layer GMM describing spatial attributes of local features contained in an image, the second-layer GMM describing intensity variation attributes of local features contained in the image.