Apparatus, method, and medium for detecting face in image using boost algorithm

• US 20060126938A1
• Filed: 07/01/2005
• Published: 06/15/2006
• Est. Priority Date: 12/09/2004
• Status: Active Grant

• Current Assignee: YP Lee MOCK & Partners
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: Kim, Jungbae, Lee, Sujin
• Granted Patent: US 7,835,541 B2
• Time in Patent Office: Days
• US Class Current: 382/190
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation

Abstract

An apparatus, method, and medium for detecting a face in an image using classifiers generated by a boost algorithm are provided. The apparatus may include an image dividing unit dividing a black and white image into sub-windows; a sub-window detection unit detecting a sub-window corresponding to a face image in the divided sub-windows using the front part of the cascade of the classifiers and storing the sub-window detected as the face image; and a post-processing unit verifying whether each of sub-windows included in a group among a plurality of groups into which stored sub-windows are grouped is the face image using the number of child detections in each group and the latter part of the cascade of classifiers, integrating the sub-windows in the group according to the result of the verification, and outputting the result of the integration as the face image included in the black and white image, wherein the number of child detections is the number of sub-windows included in a group. Accordingly, a false detection rate can be decreased while a detection rate is increased. As a result, the accuracy of face detection can be increased.

Claims

1. An apparatus for detecting a face in an image using classifiers generated by a boost algorithm, the apparatus comprising:

an image dividing unit dividing a black and white image into sub-windows;

a sub-window detection unit detecting a sub-window corresponding to a face image in the divided sub-windows using a front part of a cascade of the classifiers and storing the sub-window detected as the face image; and

a post-processing unit verifying whether each of sub-windows included in a group among a plurality of groups into which stored sub-windows are grouped is the face image using the number of child detections in each group and a latter part of the cascade of the classifiers, integrating the sub-windows in the group according to the result of the verification, and outputting the result of the integration as the face image included in the black and white image,

wherein the number of child detections is the number of sub-windows included in a group.

2. The apparatus of claim 1, further comprising an integral image computing unit generating an integral image of the black and white image and outputting the integral image to the image dividing unit,

wherein the image dividing unit divides the integral image into the sub-windows.

3. The apparatus of claim 1, further comprising a face storage unit storing at least one of a position and a size of the face image received from the post-processing unit.

4. The apparatus of claim 1, wherein the sub-window detection unit comprises:

a first window selector selecting a sub-window from among the divided sub-windows;

a first feature extractor extracting a feature of the selected sub-window;

first through n₂-th classifiers which are the front part of the cascade of the classifiers where n₂ is less than a total number of the classifiers; and

a window storage unit storing the selected sub-window in response to the result of inspection performed by the n₂-th classifier,

an x-th classifier inspects whether the selected sub-window is the face image using the feature extracted by the first feature extractor where 1≦

×

≦

n₂, and

the first window selector selects another sub-window from among the divided sub-windows in response to the result of the inspection performed by the first through n₂-th classifiers.

5. The apparatus of claim 4, wherein the sub-window detection unit further comprises:

a variance calculator calculating a variance of the selected sub-window; and

a variance comparator comparing the calculated variance with a predetermined value, and

the first feature extractor extracts the feature of the selected sub-window in response to the result of the comparison received from the variance comparator.

6. The apparatus of claim 5, wherein the predetermined value is 400.

7. The apparatus of claim 5, wherein the variance calculator calculates the variance using the following equation:

8. The apparatus of claim 4, wherein n₂ is determined according to the following condition:

9. The apparatus of claim 4, wherein n₂ is determined according to the following condition:

10. The apparatus of claim 4, wherein the post-processing unit comprises:

a grouping unit grouping the stored sub-windows into the plurality of groups and selecting and outputting the plurality of groups one by one in response to a control signal;

a counter counting the number of child detections of the selected group;

a first count comparator comparing the number of child detections with a first threshold;

a second count comparator comparing the number of child detections with a second threshold in response to the result of the comparison received from the first count comparator;

a face verifier verifying whether each of the sub-windows included in the selected group is the face image using the latter part of the cascade of classifiers in response to the result of the comparison received from the second count comparator and outputting the result of the verification;

a count adder adding the number of child detections and the number of sub-windows verified as the face image by the face verifier among the sub-windows included in the selected group;

a third count comparator comparing the result of the addition with the second threshold;

an integrator integrating the sub-windows included in the selected group in response to the result of the comparison received from one of the second and third count comparators and outputting the result of the integration as the face image included in the black and white image; and

a controller generating the control signal, and

the second threshold is greater than the first threshold.

11. The apparatus of claim 10, wherein the face verifier comprises:

a second window selector selecting a sub-window from among the sub-windows included in the selected group in response to the result of the comparison received from the second count comparator;

a second feature extractor extracting a feature of the sub-window selected by the second window selector;

n′

₁-th through N-th classifiers which are the latter part of the cascade of the classifiers where N is the total number of the classifiers; and

a counter performing a counting operation in response to the result of inspection performed by the N-th classifier and outputting the result of the counting operation to the count adder,

a y-th classifier inspects whether the sub-window selected by the second window selector is the face image using the feature extracted by the second feature extractor where n′

₁≦

y≦

N, and

the second window selector selects another sub-window from among the sub-windows included in the selected group in response to the result of the inspection performed by the n′

₁-th through N-th classifiers.

12. The apparatus of claim 10, wherein the first threshold and the second threshold are proportional to a size of the sub-window and are less than the average number of child detections, and

the average number of child detections is defined by:

$D_A=\frac{\underset{k=1}{\overset{K}{\sum <br><br>}}<br><br>\text{\hspace{1em}<br><br>}<br><br>m_k}{K},$

where D_A is the average number of child detections, K is a total number of the groups, and m_k is the number of sub-windows included in a k-th group among the K groups.

13. The apparatus of claim 11, wherein n′

₁ is determined according to the following condition;

14. The apparatus of claim 11, wherein the second feature extractor symmetrically moves the feature extracted by the first feature extractor in one of a horizontal direction and a vertical direction and outputs the result of the move as the feature of the sub-window selected by the second window selector.

15. The apparatus of claim 14, wherein the front part of the cascade of the classifiers partially overlap the latter part of the cascade of the classifiers.

16. A method of detecting a face in an image using classifiers generated by a boost algorithm, the method comprising the operations of:

(a) dividing a black and white image into sub-windows;

(b) detecting sub-windows corresponding to a face image in the divided sub-windows using a front part of a cascade of the classifiers and storing the sub-windows detected as the face image; and

(c) verifying whether each of sub-windows included in a group selected from among the plurality of groups into which the stored sub-windows are grouped is the face image using the number of child detections in each group and a latter part of the cascade of the classifiers, integrating the sub-windows in the group according to the result of the verification, and determining the result of the integration as the face image included in the black and white image,

wherein the number of child detections is the number of sub-windows included in a group.

17. The method of claim 16, further comprising:

generating an integral image of the black and white image,

wherein the operation (a) comprises dividing the integral image into the sub-windows.

18. The method of claim 16, further comprising storing at least one of a position and a size of the face image resulting from the integration.

19. The method of claim 16, wherein the operation (b) comprises:

(b1) selecting a sub-window from among the divided sub-windows;

(b2) extracting a feature of the selected sub-window;

(b3) determining whether the selected sub-window is the face image based on the extracted feature in first through n₂-th classifiers which are the front part of the cascade of the classifiers where n₂ is less than a total number of the classifiers;

(b4) when it is determined that the selected sub-window is the face image, storing the selected sub-window; and

(b5) determining whether an unselected sub-window is present in the divided sub-windows, performing (b1) when it is determined that an unselected sub-window is present, and performing (c) when it is determined that no unselected sub-window is present.

20. The method of claim 19, wherein the operation (b) further comprises:

obtaining a variance of the selected sub-window;

determining whether the variance is greater than a predetermined value; and

when it is determined that the variance is greater than the predetermined value, extracting the feature of the selected sub-window.

21. The method of claim 19, wherein the operation (c) comprises:

(c1) grouping the stored sub-windows into the plurality of groups and setting “

k”

to 1;

(c2) obtaining the number of child detections of the k-th group;

(c3) determining whether the number of child detections of the k-th group is at least a first threshold;

(c4) when it is determined that the number of child detections of the k-th group is at least the first threshold, determining whether the number of child detections of the k-th group is at least a second threshold;

(c5) when it is determined that the number of child detections of the k-th group is less than the second threshold, verifying whether each of the sub-windows included in the k-th group is the face image using the latter part of the cascade of the classifiers;

(c6) adding the number of child detections and the number of sub-windows verified as the face image among the sub-windows included in the k-th group;

(c7) determining whether the result of the addition is at least the second threshold;

(c8) when it is determined that one of the number of child detections of the k-th group and the result of the addition is at least the second threshold, integrating the sub-windows included in the k-th group and determining the result of the integration as the face image of the k-th group included in the black and white image;

(c9) determining whether “

k”

is equal to K where K is a total number of the groups; and

(c10) when it is determined that “

k”

is not equal to K, increasing “

k”

by 1 and performing (c3).

22. The method of claim 21, wherein (c5) comprises:

(c51) when it is determined that the number of child detections of the k-th group is less than the second threshold, selecting a sub-window from among the sub-windows included in the k-th group;

(c52) extracting a feature of the sub-window selected in (c51);

(c53) determining whether the sub-window selected in (c51) is the face image based on the feature extracted in (c52) in n′

₁-th through N-th classifiers which are the latter part of the cascade of classifiers where N is the total number of the classifiers;

(c54) when it is determined that the sub-window selected in (c51) is the face image, performing a counting operation; and

(c55) determining whether an unselected sub-window is present in the sub-windows included in the k-th group,

when it is determined that an unselected sub-window is present, another sub-window is selected in (c51), and

when it is determined that no unselected sub-window is present, the result of the counting operation is used as the number of sub-windows verified as the face image in (c6).

23. The method of claim 22, wherein (c52) comprises:

symmetrically moving the feature extracted in (b2) in one of a horizontal direction and a vertical direction; and

determining the result of the move as the feature of the sub-window selected in (c51).

24. At least one computer readable medium for storing instructions that control at least one processor which executes a method of detecting a face in an image using classifiers generated by a boost algorithm, wherein the method comprises:

(a) dividing a black and white image into sub-windows;

(b) detecting sub-windows corresponding to a face image in the divided sub-windows using a front part of a cascade of the classifiers and storing the sub-windows detected as the face image; and

(c) verifying whether each of sub-windows included in a group selected from among the plurality of groups into which the stored sub-windows are grouped is the face image using the number of child detections in each group and a latter part of the cascade of the classifiers, integrating the sub-windows in the group according to the result of the verification, and determining the result of the integration as the face image included in the black and white image,

wherein the number of child detections is the number of sub-windows included in a group.

25. The medium of claim 24, further comprising:

generating an integral image of the black and white image,

wherein the operation (a) comprises dividing the integral image into the sub-windows.

26. The medium of claim 24, further comprising storing at least one of a position and a size of the face image resulting from the integration.

27. The medium of claim 24, wherein the operation (b) comprises:

(b1) selecting a sub-window from among the divided sub-windows;

(b2) extracting a feature of the selected sub-window;

(b3) determining whether the selected sub-window is the face image based on the extracted feature in first through n₂-th classifiers which are the front part of the cascade of the classifiers where n₂ is less than a total number of the classifiers;

(b4) when it is determined that the selected sub-window is the face image, storing the selected sub-window; and

(b5) determining whether an unselected sub-window is present in the divided sub-windows, performing (b1) when it is determined that an unselected sub-window is present, and performing (c) when it is determined that no unselected sub-window is present.