Objectionable content detector

• US 9,684,851 B2
• Filed: 11/16/2016
• Issued: 06/20/2017
• Est. Priority Date: 04/24/2015
• Status: Active Grant

• Current Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Original Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Inventors: Abdulkader, Ahmad Abdulmageed Mohammed, Rajaram, Giridhar
• Primary Examiner(s): Carter, Aaron W
• Application Number:
  US15/353,102
• Publication Number:
  US 20170068871A1
• Time in Patent Office: 216 Days
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06F 18/254 of classification results, e.g. of results related to same input data
• CPC Class Codes: G06V 10/26 Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
• CPC Class Codes: G06V 10/44 Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
• CPC Class Codes: G06V 10/56 relating to colour
• CPC Class Codes: G06V 10/809 of classification results, e.g. where the classifiers operate on the same input data
• CPC Class Codes: G06V 30/194 References adjustable by an adaptive method, e.g. learning

Abstract

A arbitrated image classifier can be trained to identify whether an image contains specified features, such as sexual, violent, or other potentially objectionable content. An arbitrated image classifier can include a configuration of classifiers and an arbitrator that determines a final image classification based on classification results from the classifiers. An arbitrated image classifier can be trained to identify image features by dividing images labeled as including or not including a specified feature into portions, which are provided to the classifiers of the arbitrated image classifier. The arbitrator of the arbitrated image classifier can determine a result for whether or not the image includes the specified feature. If the final result does not match the image label, parameter values for various of the classifiers or the arbitrator combining procedure can be adjusted. A trained arbitrated image classifier can then be used to determine whether new images include the particular feature.

Claims

1. A method, comprising:

splitting an image associated with a label into a set of patches;

matching each selected patch, from the set of patches, to one of multiple classifiers by identifying an area of the image from which the selected patch originated and matching the identified area to an image zone;

generating, by the matched classifiers, classification results, wherein individual ones of the classification results correspond to a patch from the set of patches; and

computing a classification for the image, wherein the classification is based on one or more of the classification results.

2. The method of claim 1, further comprising:

determining that the classification does not match the label; and

in response to determining that the classification does not match the label, updating classification parameters.

3. The method of claim 1, further comprising:

providing the classification results to an arbitrator; and

wherein the classification is computed by the arbitrator.

4. The method of claim 1, further comprising an arbitrated image classifier comprising a configuration of the multiple classifiers and an arbitrator.

5. The method of claim 1, wherein the one of multiple classifiers includes a deep neural network.

6. The method of claim 1, wherein the one of multiple classifiers includes a support vector machine.

7. The method of claim 1, wherein splitting the image into patches is performed by applying an established patch template.

8. The method of claim 1, wherein splitting the image into patches is performed using a randomization algorithm.

9. The method of claim 1,

wherein each of the multiple classifiers is associated with an image zone among a plurality of image zones;

wherein matching each selected patch, from the set of patches, to one of the multiple classifiers comprises;

determining an area of the image from which the selected patch originated; and

matching the area to one image zone of a plurality of image zones;

wherein each classification result is associated with a weight value based on the image zone associated with a classifier that generated that classification result; and

wherein computing the classification is based on the weight values associated with the classification results.

10. The method of claim 1,

wherein any of the classification results are associated with a weight value;

and

wherein computing the classification uses an algorithm that accounts for the weight values associated with the classification results.

11. The method of claim 1, wherein the label indicates that the image contains sexual content.

12. The method of claim 1,

wherein the label indicates that the image contains sexual content, and

wherein none of the classification results indicate that the corresponding patch contains sexual content.

13. The method of claim 1, further comprising:

determining that the classification does not match the label; and

in response to determining that the classification does not match the label, updating classification parameters, wherein updating the classification parameters comprises one or more of;

adjusting one or more weight values associated with individual ones of the multiple classifiers;

adjusting internal parameters of one or more of the multiple classifiers;

modifying the image zone associated with one or more of the multiple classifiers;

modifying an established patch template;

or

any combination thereof.

14. The method of claim 1, wherein the image is a frame from a video file.

15. A non-transitory computer-readable storage memory storing computer-executable instructions comprising:

splitting an image into a set of patches;

matching each selected patch, from the set of patches, to one of the multiple classifiers by identifying an area of the image from which the selected patch originated and matching the identified area to an image zone;

generating, by the matched classifiers, classification results for an image feature, wherein individual ones of the classification results correspond to a patch from the set of patches; and

computing a classification for the image to identify whether the image contains the image feature, wherein the classification is based on one or more of the classification results.

16. The non-transitory computer-readable storage memory of claim 15, further comprising:

providing the classification to a user.

17. The non-transitory computer-readable storage memory of claim 15,

wherein each of the multiple classifiers is associated with an image zone among a plurality of image zones;

wherein matching each selected patch, from the set of patches, to one of the multiple classifiers comprises;

determining an area of the image from which the selected patch originated; and

matching the area to one image zone of a plurality of image zones;

wherein each classification result is associated with a weight value; and

wherein computing the classification uses an algorithm that accounts for the weight values associated with the classification results.

18. The non-transitory computer-readable storage memory of claim 15, wherein each of the multiple classifiers is a deep neural network.

19. The non-transitory computer-readable storage memory of claim 15, wherein splitting the image into patches is performed by applying an established patch template.

20. The non-transitory computer-readable storage memory of claim 15, wherein the image feature is whether or not sexual content is present.

21. A system comprising:

a memory and one or more processors; and

a component configured to;

split an image associated with a label into a set of patches;

match each selected patch, from the set of patches, to one of multiple classifiers by identifying an area of the image from which the selected patch originated and matching the identified area to an image zone;

generate classification results, wherein individual ones of the classification results correspond to a patch from the set of patches; and

compute a classification for the image, wherein the classification is based on one or more of the classification results.

22. The system of claim 21, the component further configured to:

determine that the classification does not match the label; and

in response to determining that the classification does not match the label, update;

weights associated with classification results from at least one of the classifiers of an arbitrated image classifier; and

internal features of one or more of the classifiers used to create classification results by those one or more classifiers.

23. The system of claim 21, further comprising:

a network interface configured to;

obtain an arbitrated image classifier, the arbitrated image classifier comprising a configuration of multiple classifiers and an arbitrator, wherein each of the multiple classifiers in the configuration is associated with the image zone; and

receive the image associated with the label.

24. The system of claim 21,

wherein the component uses an arbitrator to compute the classification by applying an algorithm that accounts for the weights associated with a classification results.

25. The system of claim 21, wherein the image is a frame from a video file.

26. The system of claim 21,

wherein each classification result is associated with a weight value based on the image zone associated with a classifier that generated that classification result;

and

wherein computing the classification accounts for the weight values associated with the classification results.