Image processing methods and apparatus for detecting human eyes, human face, and other objects in an image

• US 20060018517A1
• Filed: 09/27/2005
• Published: 01/26/2006
• Est. Priority Date: 09/15/2000
• Status: Active Grant

• Current Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Original Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Inventors: Ji, Xin, Chen, Xinwu
• Granted Patent: US 7,103,218 B2
• Time in Patent Office: Days
• US Class Current: 382/115
• CPC Class Codes: G06T 2207/20132 Image cropping
• CPC Class Codes: G06T 2207/30201 Face
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation

Abstract

For a subset of pixels in an image in which it is desired to detect a human face if one is present, a first variable is derived from the gray-level distribution of the image, and a second variable is derived from a preset reference distribution that is characteristic of the object. The correspondence between the first variable and the second variable is then evaluated over the subset of pixels, and a determination is made as to whether the image contains the object, based on the result of this evaluation.

Claims

1. A method for detecting an object in an image having a gray-level distribution, comprising the steps of:

a) for a subset of pixels in the image, deriving a first variable from the gray-level distribution of the image,

b) for that subset of pixels, deriving a second variable from a preset reference distribution, the reference distribution being characteristic of the subject;

c) evaluating the correspondence between the first variable and the second variable over the subset of pixels; and

d) determining whether the image contains the object, based on the result of said evaluation step.

2. The method of claim 1, further comprising the step of selecting an area in the image,

wherein pixels of the subset are in that area, and

wherein, over pixels in that area the first variable as derived from the gray-level distribution corresponding to the object to be detected has good correspondence to the second variable.

3. The method of claim 2, further comprising the steps of:

determining a sub-image in the image; and

selecting the area on the basis of the location of the sub-image.

4. The method of any one of claims 1-3, wherein the first variable presents the direction of the gradient of the gray-level distribution of the image.

5. The method of claim 4, wherein the second variable represents the direction of the gradient of the reference distribution.

6. The method of claim 5, wherein the gradient is a discrete gradient.

7. The method of claim 5, wherein said evaluation step includes a statistical process performed over the subset of pixels.

8. The method of claim 5, wherein said evaluation step includes a statistical processing over the subset of pixels and the statistical process comprises a weighting process, in which the weight assigned to each pixel in the subset is determined on the basis of the gradient of the gray level distribution at the pixel and pixels with larger magnitude of the gradient of the gray level distribution generally have larger weights.

9. The method of claim 8, wherein said evaluating step includes both the statistical process using the weighting process and another statistical process without the weighting process and determines whether the image contains the object in accordance with the results of both statistical processes.

10. The method of claim 3, wherein the area is an annular area, which centers around the center of the sub-image.

11. The method of claim 2, wherein the subset of pixel includes all the pixels in the area.

12. The method of claim 3, wherein the sub-image is determined by detecting a predetermined feature in the image.

13. The image of claim 12, wherein the predetermined feature includes a pair of dark areas.

14. The method of claim 13, wherein the sub-image is determined based on the positions of and interval between the centers of the pair of dark areas.

15. The method of claim 13, wherein the reference distribution is expressed by

z(x,y)=−

(x²/a²+y²/b²)+h

where h is a constant, a/b equals the ratio of the width of the sub-image to the height of the sub-image, and the origin is at the center of the sub-image.

16. The method of claim 5 or 15, wherein said evaluating step includes:

for all the pixels in the subset, calculating the average value of the angle between the gradient of the gray-level distribution of the image and the gradient of the reference distribution;

comparing the average value with a predetermined value; and

determining that the sub-image contains the object if the average value is equal to or smaller than the predetermined value.

17. The method of claim 5 or 15, wherein said evaluating step includes:

for all the pixels in the subset, calculating a weighted average value of the angle between the gradient of the gray-level distribution of the image and the gradient of the reference distribution, wherein a pixel having a greater magnitude of gradient of gray-level distribution generally has a larger weight;

comparing the weighted average value with a predetermined value; and

determining that the sub-image contains the object if the weighted average value is equal to or smaller than the predetermined value.

18. The method of claim 5 or 15, wherein said evaluating step includes:

for all the pixels in the subset, calculating the average value of the angle between the gradient of the gray-level distribution of the image and the gradient of the reference distribution;

for all the pixels in the subset, calculating a weighted average value of the angle between the gradient of the gray-level distribution of the image and gradient of the reference distribution, wherein a pixel having a greater magnitude of gradient of gray-level distribution has a larger weight;

comparing the average value with a first predetermined value;

comparing the weighted average value with a second predetermined value; and

determining that the sub-image contains said object if the average value is equal to or is smaller than the first predetermined value and the weighted average value is equal to or smaller than the second predetermined value.

19. A method for detecting an object in an image having a gray-level distribution, comprising the steps of:

a) determining a sub-image in the image;

b) selecting a subset of pixels in the image based on the sub-image;

c) for pixels of the subset, deriving a first variable from the gray-level distribution of the image;

d) for the subset of pixels, deriving a second variable from a preset reference distribution, the reference distribution being characteristic of the object;

e) evaluating the correspondence between the first variable and the second variable over the subset of pixels; and

f) determining whether the image contains the object based on the result of said evaluation step.

20. A computer-readable storage medium with program code stored therein for detecting an object in an image with a gray-level distribution, said program code comprising:

codes for determining a sub-image in the image;

codes for deriving, for a subset of pixels, a first variable from the gray-level distribution of the image,

codes for deriving, for the subset of pixels, a second variable from a preset reference distribution, the reference distribution being characteristic of the object;

codes for evaluating the correspondence between the first variable and the second variable over the subset of pixels; and

codes for determining whether the image contains the object based on the result of execution of said codes for evaluating.

21. An apparatus for detecting an object in an image having a gray-level distribution, comprising:

a sub-image processor adapted to determine a subset of pixels in the image;

a first calculation unit, adapted to derive a first variable from the gray-level distribution for the subset of pixels;

a second calculation unit, adapted to derive a second variable from a preset reference distribution for the subset of pixels, the reference distribution being characteristic of the object;

a correspondence evaluation processor, adapted to evaluate the correspondence between the first variable and the second variable over the subset of pixels; and

a determination unit, adapted to make a determination as to whether the image contains the object based on the evaluation result produced by said correspondence evaluation unit.