Face detection

• US 9,852,492 B2
• Filed: 09/18/2015
• Issued: 12/26/2017
• Est. Priority Date: 09/18/2015
• Status: Active Grant

• Current Assignee: Verizon Patent and Licensing Incorporated (Verizon Communications Inc.)
• Original Assignee: Yahoo Holdings, Inc. (Verizon Communications Inc.)
• Inventors: Saberian, Mohammad, Farfade, Sachin Sudhakar, Li, Jia
• Primary Examiner(s): MARIAM, DANIEL G
• Application Number:
  US14/859,040
• Publication Number:
  US 20170083752A1
• Time in Patent Office: 830 Days
• Field of Search: 382156-159, 382118
• CPC Class Codes: G06F 18/24137 Distances to cluster centroïds
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06T 3/40 Scaling of whole images or parts thereof, e.g. expanding or contracting
• CPC Class Codes: G06V 10/454 Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/82 using neural networks
• CPC Class Codes: G06V 40/165 using facial parts and geometric relationships

Abstract

Briefly, embodiments of methods and/or systems of detecting and image of a human face in a digital image are disclosed. For one embodiment, as an example, parameters of a neural network may be developed to generate object labels for digital images. The developed parameters may be refined by a neural network to generate signal sample value levels corresponding to probability that a human face may be depicted at a localized region of a digital image.

Claims

1. A method, comprising:

training parameters of a convolutional neural network (CNN) to label everyday objects; and

refining the parameters of the CNN, trained to label the everyday objects, to perform multi-aspect facial detection with respect to one or more images, the CNN being modified to exclude fully-connected layers.

2. The method of claim 1, wherein multi-aspect facial images comprise:

off-axis facial images, partially occluded facial images, cropped facial images, tilted facial images, or any combination thereof.

3. The method of claim 1, further comprising training, prior to refining and modifying to exclude fully-connected layers, the CNN utilizing a set of training input image sample values depicting the everyday objects.

4. The method of claim 3, further comprising:

modifying the set of training input image sample values to approach a threshold percentage of positive facial samples.

5. The method of claim 4, wherein a threshold percentage of positive facial samples approaches a value of approximately 20.0% to 50.0%.

6. The method of claim 1, further comprising:

executing the refining until a facial detection percentage is achieved.

7. The method of claim 1, further comprising:

scaling signal sample values corresponding to an input image to at least approximately correspond with a detector window size.

8. The method of claim 1, further comprising:

back projecting one or more output signal sample values of a subsequent convolutional filtering layer of the CNN to an original image.

9. The method of claim 1, further comprising:

displaying a heat map corresponding to one or more localized regions at which a depiction of one or more human faces is likely to be present in an input image.

10. The method of claim 1, wherein the CNN comprises an eight-layer deep CNN.

11. The method of claim 1, wherein the modifying of the CNN comprises excluding three fully-connected layers.

12. An apparatus, comprising:

one or more processors to;

train parameters of a convolutional neural network (CNN) to label everyday objects; and

refine parameters of the CNN, trained to label the everyday objects, to perform multi-aspect facial detection with respect to one or more images, the CNN modified to exclude fully-connected layers.

13. The apparatus of claim 12, wherein multi-aspect facial images comprise off-axis facial images, partially occluded facial images, cropped facial images, tilted facial images, or any combination thereof.

14. The apparatus of claim 12, wherein the one or more processors are additionally to:

train the CNN to classify input image signal samples depicting the everyday objects before the CNN is modified to exclude fully-connected layers.

15. The apparatus of claim 12, wherein the one or more processors are additionally to:

scale signal sample values corresponding to an input image to at least approximately correspond with a detector window size.

16. The apparatus of claim 12, wherein the one or more processors are additionally to:

back project one or more output signal samples of a subsequent convolutional filtering of the CNN to an original image.

17. An apparatus comprising:

means for training parameters of a convolutional neural network (CNN) to label everyday objects; and

means for refining the parameters of the CNN, trained to label the everyday objects, to perform multi-aspect facial detection, wherein the means for refining comprises means for modifying the CNN to exclude fully-connected layers.

18. The apparatus of claim 17, wherein the means for refining the CNN to perform multi-aspect facial detection comprises means for detecting off-axis facial images, partially occluded facial images, cropped facial images, tilted facial images, or any combination thereof.

19. The apparatus of claim 18, further comprising:

means for scaling signal sample values corresponding to an input image to at least approximately correspond with a detector window size.

20. The apparatus of claim 18, further comprising:

means for back projecting one or more output signal sample values of a subsequent convolutional filtering layer of the CNN to an original image.