Face Detection Using Division-Generated Haar-Like Features For Illumination Invariance

• US 20120019683A1
• Filed: 07/26/2010
• Published: 01/26/2012
• Est. Priority Date: 07/26/2010
• Status: Active Grant

• Current Assignee: FotoNation Limited (Adeia, Inc.)
• Original Assignee: FotoNation Limited (Adeia, Inc.)
• Inventors: Susanu, George, Gangea, Mihnea, Filip, Dan
• Granted Patent: US 8,971,628 B2
• Time in Patent Office: Days
• US Class Current: 348/222.1
• CPC Class Codes: G06V 10/50 by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation

Abstract

Faces in images are quickly detected with minimal memory resource usage. Instead of calculating a Haar-like feature value by subtracting the average pixel intensity value in one rectangular region from the average pixel intensity value in another, adjacent rectangular region, a face-detection system calculates that Haar-like feature value by dividing the average pixel intensity value in one such rectangular region by the average pixel intensity value in the other such adjacent rectangular region. Thus, each Haar-like value is calculated as a ratio of average pixel intensity values rather than as a difference between such average pixel intensity values. The feature values are calculated using this ratio-based technique both during the machine-learning procedure, in which the numerical ranges for features in known face-containing images are learned based on labeled training data, and during the classifier-applying procedure, in which an unlabeled image'"'"'s feature values are calculated and compared to the previously machine-learned numerical ranges.

Claims

1. A method comprising particular steps of:

determining a first intensity value for a first region of an image;

determining a second intensity value for a second region of the image;

generating a first feature value for a first feature at least in part by dividing the first intensity value by the second intensity value;

determining whether the first feature value falls within a specified range that is associated with the first feature; and

in response to determining that the first feature value falls within the specified range that is associated with the first feature, storing data that indicates that the image contains a face;

wherein the steps are performed by an automated device.

2. The method of claim 1, wherein the first region is a rectangular region, wherein the second region is a rectangular region, and wherein the first region is adjacent to the second region.

3. The method of claim 1, wherein generating the first feature value comprises dividing the first intensity value by the second intensity value and normalizing a quotient produced by dividing the first intensity value by the second intensity value, and wherein normalizing the quotient comprises multiplying the quotient by a specified value.

4. The method of claim 1, wherein generating the first feature value is performed without using any statistical variance of pixel intensity values in the image.

5. The method of claim 1, wherein the image is a source image, and wherein generating the first feature value is performed without using any integral image that was generated based on the source image.

6. The method of claim 1, further comprising:

determining a third intensity value for a third region of the image;

determining a fourth intensity value for a fourth region of the image;

generating a second feature value for a second feature at least in part by dividing the third intensity value by the fourth intensity value;

determining whether the second feature value falls within a specified range that is associated with the second feature;

wherein the particular steps are performed in response to determining that the second feature value falls within the specified range that is associated with the second feature;

wherein the third region is neither the first region nor the second region;

wherein the fourth region is neither the first region nor the second region;

wherein the third region is adjacent to the fourth region.

7. The method of claim 1, wherein determining the first intensity value comprises determining an average of two or more intensity values of pixels that are contained within the first region, and wherein determining the second intensity value comprises determining an average of two or more intensity values of pixels that are contained within the second region.

8. The method of claim 1, wherein the specified range is determined automatically based on a machine-learning technique that evaluates both (a) images that are known to contain a face and (b) images that are known not to contain a face.

12. The computer-readable non-transitory storage medium of claim 1, wherein the first region is a rectangular region, wherein the second region is a rectangular region, and wherein the first region is adjacent to the second region.

13. The computer-readable non-transitory storage medium of claim 1, wherein generating the first feature value comprises dividing the first intensity value by the second intensity value and normalizing a quotient produced by dividing the first intensity value by the second intensity value, and wherein normalizing the quotient comprises multiplying the quotient by a specified value.

14. The computer-readable non-transitory storage medium of claim 1, wherein generating the first feature value is performed without using any statistical variance of pixel intensity values in the image.

15. The computer-readable non-transitory storage medium of claim 1, wherein the image is a source image, and wherein generating the first feature value is performed without using any integral image that was generated based on the source image.

16. The computer-readable non-transitory storage medium of claim 1, wherein steps performed by the one or more processors further comprise:

determining a third intensity value for a third region of the image;

determining a fourth intensity value for a fourth region of the image;

generating a second feature value for a second feature at least in part by dividing the third intensity value by the fourth intensity value;

determining whether the second feature value falls within a specified range that is associated with the second feature;

wherein the particular steps are performed in response to determining that the second feature value falls within the specified range that is associated with the second feature;

wherein the third region is neither the first region nor the second region;

wherein the fourth region is neither the first region nor the second region;

wherein the third region is adjacent to the fourth region.

17. The computer-readable non-transitory storage medium of claim 1, wherein determining the first intensity value comprises determining an average of two or more intensity values of pixels that are contained within the first region, and wherein determining the second intensity value comprises determining an average of two or more intensity values of pixels that are contained within the second region.

18. The computer-readable non-transitory storage medium of claim 1, wherein the specified range is determined automatically based on a machine-learning technique that evaluates both (a) images that are known to contain a face and (b) images that are known not to contain a face.

9. A method comprising:

generating a look-up table that associates key values with corresponding reciprocal values;

determining a first intensity value for a first region of an image;

determining a second intensity value for a second region of the image;

generating a first feature value for a first feature at least in part by multiplying the first intensity value by a reciprocal value that is associated with the second intensity value in the look-up table;

determining whether the first feature value falls within a specified range that is associated with the first feature; and

in response to determining that the first feature value falls within the specified range that is associated with the first feature, storing data that indicates that the image contains a face;

wherein the steps are performed by an automated device.

10. The method of claim 9, wherein the look-up table contains 256 entries.

11. A computer-readable non-transitory storage medium storing instructions which, when executed by one or more processors, cause the processors to perform particular steps comprising:

determining a first intensity value for a first region of an image;

determining a second intensity value for a second region of the image;

generating a first feature value for a first feature at least in part by dividing the first intensity value by the second intensity value;

determining whether the first feature value falls within a specified range that is associated with the first feature; and

in response to determining that the first feature value falls within the specified range that is associated with the first feature, storing data that indicates that the image contains a face.

19. A computer-readable non-transitory storage medium storing instructions which, when executed by one or more processors, cause the processors to perform steps comprising:

generating a look-up table that associates key values with corresponding reciprocal values;

determining a first intensity value for a first region of an image;

determining a second intensity value for a second region of the image;

generating a first feature value for a first feature at least in part by multiplying the first intensity value by a reciprocal value that is associated with the second intensity value in the look-up table;

determining whether the first feature value falls within a specified range that is associated with the first feature; and

in response to determining that the first feature value falls within the specified range that is associated with the first feature, storing data that indicates that the image contains a face.

20. The computer-readable non-transitory storage medium of claim 19, wherein the look-up table contains 256 entries.

21. A digital camera comprising:

an image capturing and storing module that captures and stores an image;

an intensity value-determining module that determines (a) a first intensity value for a first region of an image and (b) a second intensity value for a second region of the image;

a feature value-generating module that generates a first feature value for a first feature at least in part by dividing the first intensity value by the second intensity value;

a comparator that determines whether the first feature value falls within a specified range that is associated with the first feature; and

a memory module which, in response to a determination that the first feature value falls within the specified range that is associated with the first feature, stores data that indicates that the image contains a face.

22. The digital camera of claim 21, wherein the first region is a rectangular region, wherein the second region is a rectangular region, and wherein the first region is adjacent to the second region.

23. The digital camera of claim 21, wherein, to generate the first feature value, the feature value-generating module divides the first intensity value by the second intensity value and normalizes a quotient produced by dividing the first intensity value by the second intensity value, and wherein to normalize the quotient value, the feature value-generating module multiplies the quotient by a specified value.

24. The digital camera of claim 21, wherein, while generating the first feature value, the feature value-generating module does not use any statistical variance of pixel intensity values in the image.

25. The digital camera of claim 21, wherein the image is a source image, and wherein, while generating the first feature value, the feature value-generating module does not use any integral image that was generated based on the source image.

26. The digital camera of claim 21, wherein the feature value-generating module determines the first intensity value at least in part by determining an average of two or more intensity values of pixels that are contained within the first region, and wherein the feature value-generating module determines the second intensity value at least in part by determining an average of two or more intensity values of pixels that are contained within the second region.

27. The digital camera of claim 21, wherein the specified range is determined automatically based on a machine-learning technique that evaluates both (a) images that are known to contain a face and (b) images that are known not to contain a face.

28. A digital camera comprising:

an image capturing and storing module that captures and stores an image;

an intensity value-determining module that determines (a) a first intensity value for a first region of an image and (b) a second intensity value for a second region of the image;

an feature value-generating module that generates a first feature value for a first feature at least in part by multiplying the first intensity value by a reciprocal value that is associated with the second intensity value in a stored look-up table that associates key values with corresponding reciprocal values;

a comparator that determines whether the first feature value falls within a specified range that is associated with the first feature; and

a memory module which, in response to a determination that the first feature value falls within the specified range that is associated with the first feature, stores data that indicates that the image contains a face.

29. The digital camera of claim 28, wherein the look-up table contains 256 entries.