TECHNIQUES FOR ENABLING OR ESTABLISHING THE USE OF FACE RECOGNITION ALGORITHMS

• US 20070258645A1
• Filed: 03/12/2007
• Published: 11/08/2007
• Est. Priority Date: 03/12/2006
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Shah, Munjal, Lee, Kuang-Chih, Torresani, Lorenzo, Vu, Diem, Vanhoucke, Vincent, Gokturk, Salih, Anguelov, Dragomir
• Granted Patent: US 8,571,272 B2
• Time in Patent Office: Days
• US Class Current: 382/190
• CPC Class Codes: G06V 10/7557 based on appearance, e.g. active appearance models [AAM]
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

Embodiments described herein facilitate or enhance the implementation of image recognition processes which can perform recognition on images to identify objects and/or faces by class or by people.

Claims

1. A computer-implemented method for performing recognition on an image, The computer-implemented method comprising:

identifying, from image data, an object that is posed so as to occlude one or more regions of the object;

synthesizing image data for at least some of the one or more occluded regions of the object;

performing recognition of the object using the synthesized image data.

2. The computer-implemented method of claim 1, wherein the object is a face of a person.

3. The computer-implemented method of claim 1, wherein synthesizing image data includes using a three-dimensional data model of the object.

4. The computer-implemented method of claim 3, wherein synthesizing image data includes augmenting the three-dimensional data model of the object with attributes of the image data.

5. The computer-implemented method of claim 4, wherein augmenting the three-dimensional data model includes posing the three-dimensional data model to correspond to a view of the object in the image data.

6. The computer-implemented method of claim 5, wherein augmenting the three-dimensional data model includes computing an intensity or value of discrete locations of the posed three-dimensional model from an intensity or value of one or more corresponding discrete locations of the object of the image data.

11. The computer-implemented method of claim 1O, wherein using images from the training set to perform a recognition includes presenting to the user a set of recognized images that each include a recognized object or face that is contained in each image of the training set, and wherein the method further comprises enabling the user to identify if any of the images in the set of recognized images includes an erroneous recognition.

12. The computer-implemented method of claim 11, programmatically selecting includes identifying a cluster of images from a collection, wherein each image in the cluster is determined to contain image data of an object or person, and wherein identifying a cluster of images includes determining that a confidence value of each image containing image data of an object or person is within a defined threshold, and wherein method further comprises tracking each instance in which the user identifies an erroneous recognition, and adjusting the defined threshold of the confidence values based on each instance in which the user identifies the erroneous recognition.

13. The computer-implemented method of claim 12, further comprising reducing or increasing the defined threshold of confidence values based on a measurement of the instances of erroneous recognition.

7. A computer-implemented method for performing recognition on an image, The computer-implemented method comprising:

programmatically selecting, from a collection of images, a set of candidate images for a training set;

presenting the set of images to a user for input;

designating images from the set of candidate images for the training set based on input from the user; and

using images from the training set to perform a recognition of one or more images.

8. The computer-implemented method of claim 7, wherein programmatically selecting includes identifying a cluster of images from a collection, wherein each image in the cluster is determined to contain an image of an object or person.

9. The computer-implemented method of claim 8, identifying a cluster of images includes determining that a confidence value of each image is within a defined threshold.

10. The computer-implemented method of claim 7, wherein using images from the training set to perform a recognition includes performing at least one of (i) presenting information about the recognition to a user for input, or (ii) presenting one or more images that have a recognized object or face that is contained in each image of the training set.

14. The computer-implemented method of claim 7, wherein programmatically selecting includes performing a simulated recognition of images in a collection.

15. A computer-implemented method for performing recognition on an image, method comprising:

determining a region of an image where an object that is to be recognized is likely located;

for each of a plurality of features of the object that is to be recognized, determining a set of one or more hypotheses regions of the image that are likely locations of that feature in the object that is to be recognized.

determining a most likely location of at least some of the plurality of features using the set of one or more hypothesis regions for all of the plurality of features.

16. The method of claim 16, wherein the object is a face, and wherein the set of additional features include one or more features selected from a set consisting of eyes, a nose tip, ends of a mouth, ear locations, hairline, and eyebrows, as well as other parts of the face.

17. The method of claim 18, wherein the predefined model is a two-dimensional or three dimensional generic model.

18. The method of claim 16, further comprising determining the plurality of features for the object using a predefined object model.

19. The method of claim 19, wherein determining a set of one or more hypotheses regions of the image that includes deriving at least some of the hypotheses locations from the predefined object model, using one or more geometric constraints that exist between a likely location of two or more of the plurality of features.

20. The method of claim 20, further comprising determining a pose of the object that is to be recognized while concurrently determining the most likely location of at least some of the plurality of features.