Object detection using dynamic probability scans

• US 6,912,298 B1
• Filed: 08/11/2003
• Issued: 06/28/2005
• Est. Priority Date: 08/11/2003
• Status: Active Grant

• Current Assignee: Adobe Inc.
• Original Assignee: Adobe Systems Incorporated (Adobe Inc.)
• Inventors: Wilensky, Gregg D.
• Primary Examiner(s): Hjerpe, Richard
• Assistant Examiner(s): Lay, Michelle K.
• Application Number:
  US10/639,238
• Time in Patent Office: 687 Days
• Field of Search: 382/194, 382/190, 382/195, 382/199, 382/117, 382/115, 345/581, 345/619
• US Class Current: 382/117
• CPC Class Codes: G06T 2207/30201 Face
• CPC Class Codes: G06T 2207/30216 Redeye defect
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling
• CPC Class Codes: G06T 7/73 using feature-based methods
• CPC Class Codes: G06T 7/90 Determination of colour characteristics
• CPC Class Codes: G06V 10/507 Summing image-intensity values; Histogram projection analysis
• CPC Class Codes: G06V 40/19 Sensors therefor

Abstract

Methods and apparatus, including computer program products, for detecting an object in an image. The techniques include scanning a sequence of pixels in the image, each pixel having one or more property values associated with properties of the pixel, and generating a dynamic probability value for each of one or more pixels in the sequence. The dynamic probability value for a given pixel represents a probability that the given pixel has neighboring pixels in the sequence that correspond to one or more features of the object. The dynamic probability value is generated by identifying a dynamic probability value associated with a pixel that immediately precedes the given pixel in the sequence; updating the identified dynamic probability value based on the property values of the immediately preceding pixel; and associating the updated probability value with the given pixel.

Claims

1. A computer-implemented method for detecting an object in an image, the method comprising:

scanning a sequence of pixels in the image, each pixel having one or more property values associated with properties of the pixel; and

generating a dynamic probability value for each of one or more pixels in the sequence, wherein the dynamic probability value for a given pixel represents a probability that the given pixel has neighboring pixels in the sequence that correspond to one or more features of the object and the dynamic probability value is generated by;

identifying a dynamic probability value associated with a pixel that immediately precedes the given pixel in the sequence;

updating, the identified dynamic probability value based on the property values of the immediately preceding pixel; and

associating the updated probability value with the given pixel.

2. The method of claim 1, wherein updating the identified dynamic probability value includes:

calculating an intrinsic probability value for the immediately preceding pixel based on one or more property values for the immediately preceding pixel, wherein the intrinsic probability value for the immediately preceding pixel represents a probability that the immediately preceding pixel corresponds to one or more features of the object; and

updating the identified dynamic probability value based on the intrinsic probability value calculated for the immediately preceding pixel.

3. The method of claim 2, wherein one or more of the property values for the immediately preceding pixel are derived from property values of other nearby pixels.

4. The method of claim 1, wherein updating the identified dynamic probability value includes:

multiplying the identified dynamic probability value by a weighting coefficient.

5. The method of claim 1, wherein:

the dynamic probability value is a vector that includes one or more components, each component corresponding to a different feature of the object.

6. The method of claim 1, further comprising:

using the dynamic probability value of a pixel to calculate an object probability value for the pixel, the object probability value representing the probability that the pixel corresponds to the object.

7. The method of claim 1, wherein:

the object is an eye; and

a feature of the object includes a redeye effect.

8. A computer-implemented method for detecting an object in an image, the method comprising:

scanning a sequence of pixels in the image, each pixel having one or more property values for one or more properties of the pixel;

calculating a set of feature probability values for each of one or more pixels in the sequence, each feature probability value representing a probability that the pixel corresponds to a feature of the object, each feature probability in the set corresponding to a different feature of the object; and

using the feature probability values to calculate an object probability value for the pixel, the object probability value representing the probability that the pixel corresponds to the object,

wherein each feature probability includes a dynamic probability value that is generated by;

identifying the dynamic probability value of an immediately preceding pixel in the sequence;

updating the identified dynamic probability value to account for the intrinsic properties of the immediately preceding pixel; and

associating the updated probability value with the pixel.

9. The method of claim 8, wherein the set of feature probability values includes:

a first feature probability value corresponding to a primary feature of the object; and

a second feature probability value corresponding to a secondary feature of the object, the primary feature being different from the secondary feature.

10. The method of claim 9, wherein:

the object is an eye; and

the primary feature of the object is a redeye effect.

11. The method of claim 9, further comprising:

comparing the object probability value against a threshold value; and

if the object probability value meets or exceeds the threshold value, then determining that the pixel corresponds to the object.

12. A computer-implemented method for detecting an object in an image, the method comprising:

identifying a set of candidate pixels in the image; and

reducing the number of pixels in the set of candidate pixels by performing a dynamic probability scan of the image, wherein performing the dynamic probability scan includes;

scanning a sequence of pixels in the image, each pixel having one or more intrinsic properties including one or more color values; and

generating a dynamic probability value for each of one or more pixels in the sequence, wherein the dynamic probability value for a given pixel represents a probability that the given pixel has neighboring pixels in the sequence that correspond to a feature of the object and the dynamic probability value is generated by;

identifying the dynamic probability value generated for a pixel that immediately precedes the given pixel in the sequence;

updating the identified dynamic probability value to account for the intrinsic properties of the immediately preceding pixel; and

associating the updated probability value with the given pixel.

13. The method of claim 12, further comprising:

identifying one or more pixels in the reduced set of candidate pixels as corresponding to the object.

14. A computer-implemented method for detecting an object in an image, the method comprising:

scanning a sequence of pixels in the image; and

calculating a dynamic probability value for a pixel in the sequence of pixels based on the following equation;

pD(x+1)=p(x)+w*pD(x),

wherein;

x represents a position in the sequence;

p(x) is an intrinsic probability value representing the probability that a pixel located at position x corresponds to a feature of the object;

pD(x) is a dynamic probability value representing the probability that a pixel located at position x has neighboring pixels in the sequence that correspond to a feature of the object; and

w is a weighting coefficient that determines how influence pD(x) has on pD(x+1).

15. A computer-program product, tangibly embodied in a computer readable medium, for detecting an object in an image, the computer program product being operable to cause data processing equipment to perform operations comprising:

scanning a sequence of pixels in the image, each pixel having one or more property values associated with properties of the pixel; and

generating a dynamic probability value for each of one or more pixels in the sequence, wherein the dynamic probability value for a given pixel represents a probability that the given pixel has neighboring pixels in the sequence that correspond to one or more features, of the object and the dynamic probability value is generated by;

identifying a dynamic probability value associated with a pixel that immediately precedes the given pixel in the sequence;

updating the identified dynamic probability value based on the property values of the immediately preceding pixel; and

associating the updated probability value with the given pixel.

16. The product of claim 15, wherein updating the identified dynamic probability value includes:

calculating an intrinsic probability value for the immediately preceding pixel based on one or more property values for the immediately preceding pixel, wherein the intrinsic probability value for the immediately preceding pixel represents a probability that the immediately preceding pixel corresponds to one or more features of the object; and

updating the identified dynamic probability value based on the intrinsic probability value calculated for the immediately preceding pixel.

17. The product of claim 16, wherein one or more of the property values for the immediately preceding pixel are derived from property values of other nearby pixels.

18. The product of claim 15, wherein updating the identified dynamic probability value includes:

multiplying the identified dynamic probability value by a weighting coefficient.

19. The product of claim 15, wherein:

the dynamic probability value is a vector that includes one or more components, each component corresponding to a different feature of the object.

20. The product of claim 15, wherein the operations further comprise:

using the dynamic probability value of a pixel to calculate an object probability value for the pixel, the object probability value representing the probability that the pixel corresponds to the object.

21. The product of claim 15, wherein:

the object is an eye; and

a feature of the object includes a redeye effect.

22. A computer-program product, tangibly embodied in a computer readable medium, for detecting an object in an image, the computer program product being operable to cause data processing equipment to perform operations comprising:

scanning a sequence of pixels in the image, each pixel having one or more property values for one or more properties of the pixel;

calculating a set of feature probability values for each of one or more pixels in the sequence, each feature probability value representing a probability that the pixel corresponds to a feature of the object, each feature probability in the set corresponding to a different feature of the object; and

using the feature probability values to calculate an object probability value for the pixel, the object probability value representing the probability that the pixel corresponds to the object,

wherein each feature probability includes a dynamic probability value that is generated by;

identifying the dynamic probability value of an immediately preceding pixel in the sequence;

updating the identified dynamic probability value to account for the intrinsic properties of the immediately preceding pixel; and

associating the updated probability value with the pixel.

23. The product of claim 22, wherein the set of feature probability values includes:

a first feature probability value corresponding to a primary feature of the object; and

a second feature probability value corresponding to a secondary feature of the object, the primary feature being different from the secondary feature.

24. The product of claim 23, wherein:

the object is an eye; and

the primary feature of the object is a redeye effect.

25. The product of claim 23, further comprising:

comparing the object probability value against a threshold value; and

if the object probability value meets or exceeds the threshold value, then determining that the pixel corresponds to the object.

26. A computer-program product, tangibly embodied in a computer readable medium, for detecting an object in an image, the computer program product being operable to cause data processing equipment to perform operations comprising:

identifying a set of candidate pixels in the image; and

reducing the number of pixels in the set of candidate pixels by performing a dynamic probability scan of the image, wherein performing the dynamic probability scan includes;

scanning a sequence of pixels in the image, each pixel having one or more intrinsic properties including one or more color values; and

generating a dynamic probability value for each of one or more pixels in the sequence, wherein the dynamic probability value for a given pixel represents a probability that the given pixel has neighboring pixels in the sequence that correspond to a feature of the object and the dynamic probability value is generated by;

identifying the dynamic probability value generated for a pixel that immediately precedes the given pixel in the sequence;

updating the identified dynamic probability value to account for the intrinsic properties of the immediately preceding pixel; and

associating the updated probability value with the given pixel.

27. The product of claim 26, further comprising:

identifying one or more pixels in the reduced set of candidate pixels as corresponding to the object.

28. A computer-program product, tangibly embodied in a computer readable medium, for detecting an object in an image, the computer program product being operable to cause data processing equipment to perform operations comprising:

scanning a sequence of pixels in the image; and

calculating a dynamic probability value for a pixel in the sequence of pixels based on the following equation;

pD(x+1)=p(x)+w*pD(x),

wherein;

x represents a position in the sequence;

p(x) is an intrinsic probability value representing the probability that a pixel located at position x corresponds to a feature of the object;

pD(x) is a dynamic probability value representing the probability that a pixel located at position x has neighboring pixels in the sequence that correspond to a feature of the object; and

w is a weighting coefficient that determines how influence pD(x) has on pD(x+1).