Face template balancing

Patent metadata

Field	Value
Publication number	US-2016283779-A1
Application number	US-201414334354-A
Country	US
Kind code	A1
Filing date	Jul 17, 2014
Priority date	Jul 19, 2013
Publication date	Sep 29, 2016
Grant date	—

Abstract

Implementations generally relate to face template balancing. In some implementations, a method includes generating face templates corresponding to respective images. The method also includes matching the images to a user based on the face templates. The method also includes receiving a determination that one or more matched images are mismatched images. The method also includes flagging one or more face templates corresponding to the one or more mismatched images as negative face templates.

First claim

1 . A computer-implemented method comprising:
generating face templates corresponding to respective images, wherein each face template describes one or more facial features of a face depicted in a corresponding one of the images and stores at least one value for at least one of the one or more facial features; matching the images to a user identity of a user by determining similarities of the one or more facial features in the face templates corresponding to the images to one or more facial features in one or more other images that depict the user; determining based on received user input that one or more of the matched images are mismatched images that do not depict the user of the user identity; flagging one or more face templates corresponding to the one or more mismatched images as negative face templates; determining based on received user input that one or more of the matched images are correctly matched images that depict the user of the user identity; flagging one or more face templates corresponding to the one or more correctly matched images as positive face templates; generating a face model associated with the user based on combining the values of the facial features of a plurality of the positive face templates, wherein the face model is configured to be compared to corresponding face templates of one or more new images to determine whether the one or more new images match the user identity; and associating the face model with at least one of the negative face templates, wherein the at least one negative face template is configured to indicate whether one or more facial features in the one or more new images are not matched with the user identity.

2 . The method of claim 1 , wherein one or more of the face templates describe location of facial features on the face and distances between facial features of the face, and wherein the combined values in the face model include averaged values for a plurality of facial features in the plurality of positive face templates.

3 . The method of claim 1 , further comprising causing a display of the matched images by a device, wherein determining that one or more of the matched images are mismatched images comprises
receiving by the device one or more indications from the user indicating that one or more of the displayed matched images are mismatched images, and wherein the determining that one or more of the matched images are correctly matched images comprises receiving by the device one or more indications from the user indicating that one or more of the displayed matched images are correctly matched images.

4 . The method of claim 1 , further comprising:
generating a new face template for a new image of the one or more new images, wherein the new face template describes one or more facial features of a face depicted in the new image and stores at least one value for at least one of the one or more facial features in the new face template; comparing one or more of the facial features in the new face template of the new image with one or more of the facial features in the face model; determining a match between the new image and the face model based on the comparing of the one or more facial features in the new face template with the one or more facial features in the face model; comparing one or more of the facial features in the new face template with one or more of the facial features in the at least one negative face template associated with the face model; and removing the determination of the match of the new image and the face model based on the comparing of the one or more facial features in the new face template with the one or more facial features in the at least one negative face template.

5 . A computer-implemented method comprising:
generating a face template corresponding to an image, wherein the face template describes one or more facial features of a face depicted in the image and stores at least one value for at least one of the one or more facial features; comparing one or more facial features in the face template with one or more facial features in a face model describing a face of a user, wherein the facial features in the face model are associated with values determined from values of facial features of a plurality of positive face templates generated from images verified to depict the user; comparing one or more facial features in the face template with one or more facial features in one or more negative face templates, wherein the one or more negative face templates are generated from one or more images having face depictions verified to not depict the user; and determining whether the image depicts the user based on the comparison with the one or more facial features in the face model and based on the comparison with the one or more facial features in the one or more negative face templates.

6 . The method of claim 5 , further comprising, prior to the comparing:
generating one or more face templates from one or more images, wherein the face template describes one or more facial features of a face depicted in the image and store at least one value for at least one of the one or more facial features; determining matches of the one or more face templates to a user identity of the user by determining similarities of one or more facial features described in the one or more face templates to one or more facial features described in one or more other images that are indicated to depict the user; causing, on a device, a display of the images corresponding to the one or more matched face templates; receiving by the device user input indicating that one or more of the displayed images are mismatched images that do not depict the user; and flagging the one or more face templates corresponding to the one or more mismatched images as the one or more negative face templates.

7 . The method of claim 5 , wherein determining whether the image depicts the user includes determining that the image does not depict the user, based on the face template corresponding to the image being more similar to at least one of the one or more negative templates than to the face model.

8 . The method of claim 5 , wherein determining whether the image depicts the user includes determining that the image depicts the user, based on the face template corresponding to the image being more similar to the face model than to the one or more negative templates.

9 . The method of claim 5 , further comprising:
causing a display of the image by a device; and receiving by the device user input indicating that the displayed matched image is a mismatched image that does not depict the user.

10 . The method of claim 8 , further comprising:
updating the face model with the face template corresponding to the image.

11 . The method of claim 5 , further comprising:
causing a display of the image by a device; and receiving by the device user input indicating that the displayed image is a correctly matched image that depicts the user.

12 . The method of claim 5 , wherein determining whether the image depicts the user includes:
determining one or more differentiating characteristics that differentiate the one or more negative face templates from the one or more positive face templates, wherein the differentiating characteristics include at least one of:
one or more facial features in the one or more negative face templates or in the one or more positive face templates, and
one or more colors provided in the face described by the facial features;
determining that the face template corresponding to the image includes one or more of the differentiating characteristics as a weight indicating that the image does not

Assignees

• Google Inc

Inventors

• Mcphie Jonathan
• Adam Hartwig
• Fredinburg Dan
• Masterov Alexei

Classifications

• G06V10/7788 Primary

  the supervisor being a human, e.g. interactive learning with a human teacher · CPC title

• G06V10/772

  Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries · CPC title

• G06V40/164

  using holistic features · CPC title

• G06F18/41

  Interactive pattern learning with a human teacher · CPC title

• G06F18/28

  Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries · CPC title