METHOD, SYSTEM, AND COMPUTER READABLE MEDIUM FOR FACE DETECTION

• US 20100172581A1
• Filed: 12/18/2009
• Published: 07/08/2010
• Est. Priority Date: 12/19/2008
• Status: Active Grant

• Current Assignee: Cisco Technology, Inc. (Cisco Systems, Inc.)
• Original Assignee: Tandberg Telecom AS (Cisco Systems, Inc.)
• Inventors: Husøy, Per Ove
• Granted Patent: US 8,260,011 B2
• Time in Patent Office: Days
• US Class Current: 382/165
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/10024 Color image
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 2207/30201 Face
• CPC Class Codes: G06T 7/90 Determination of colour characteristics
• CPC Class Codes: G06V 10/56 relating to colour
• CPC Class Codes: G06V 40/167 using comparisons between temporally consecutive images

Abstract

A method, system, and computer readable medium for detecting a face in a video stream including receiving a sequence of input color images; and for each input image, calculating a greyscale image of the input color image, creating a one-bit motion image based on the current and a previous greyscale image, calculating a normalized color image of the input color image, calculating a motion color probability image, providing at least the grayscale image and the motion color probability image to a face detector, and executing face detection by determining a presence of a face based on first features in the grayscale image and second features in the motion color probability image.

Claims

1. A method for detecting a face in a video stream comprising:

receiving, at a video processing apparatus, a sequence of input color images; and

for each input image, processing the sequence at the video processing apparatus, wherein the process includes,

calculating a greyscale image of the input color image,

creating a one-bit motion image based on the current greyscale image and a previous greyscale image,

calculating a normalized color image of the input color image,

calculating a motion color probability image by,

calculating a first color histogram of the pixels in the normalized image,

calculating a second color histogram of the pixels in the normalized image corresponding to pixels in the motion image defined as being in motion,

calculating a ratio histogram by dividing the first color histogram by the second color histogram, and

setting pixel values in the motion color probability image to a value obtained by indexing the ratio histogram using values from the original normalized image,

providing at least the grayscale image and the motion color probability image to a face detector, and

executing face detection by determining a presence of a face based on first features in the grayscale image and second features in the motion color probability image.

2. The method according to claim 1, wherein the first features include geometrical features or patterns.

3. The method according to claim 1, wherein the second features include intensity information, the intensity information defining the presence or non-presence of assumed skin colors.

4. The method according to claim 1, further comprising:

providing the motion image to the face detector, wherein the presence of a face is further based on third features in the motion image.

5. The method according to claim 1, wherein the input color image is in the RGB color space, and wherein the normalized image is created by converting a copy of the input color image to the RG chromaticity space.

6. The method according to claim 1, wherein when the face detector detects a face in one of the input images, the calculating of the motion color probability image extracts the location of the detected face and calculates a second color histogram of the pixels representing the detected face.

7. The method according to claim 1, wherein the receiving of the sequence of input color images is received from a static video camera.

8. An apparatus for detecting a face in a video stream comprising:

a video processor that receives a sequence of input color images; and

for each input image, the video processor,

calculates a greyscale image of the input color image,

creates a one-bit motion image based on the current greyscale image and a previous greyscale image,

calculates a normalized color image of the input color image,

calculates a motion color probability image, wherein the video processor further,

calculates a first color histogram of the pixels in the normalized image,

calculates a second color histogram of the pixels in the normalized image corresponding to pixels in the motion image defined as being in motion,

calculates a ratio histogram by dividing the first color histogram by the second color histogram, and

sets pixel values in the motion color probability image to a value obtained by indexing the ratio histogram using values from the original normalized image,

provides at least the grayscale image and the motion color probability image to a face detector, and

executes face detection by determining a presence of a face based on first features in the grayscale image and second features in the motion color probability image.

9. The apparatus according to claim 8, wherein the first features include geometrical features or patterns.

10. The apparatus according to claim 8, wherein the second features include intensity information, the intensity information defining the presence or non-presence of is assumed skin colors.

11. The apparatus according to claim 8, wherein the video processor further provides the motion image to the face detector, and wherein the presence of a face is further based on third features in the motion image.

12. The apparatus according to claim 8, wherein the input color image is in the RGB color space, and wherein the normalized image is created by converting a copy of the input color image to the RG chromaticity space.

13. The apparatus according to claim 8, wherein when the face detector detects a face in one of the input images, the video processor extracts the location of the detected face and calculates a second color histogram of the pixels representing the detected face.

14. The apparatus according to claim 8, wherein the video processor receives the sequence of input color images from a static video camera.

15. A computer readable storage medium encoded with computer executable instructions, wherein the instructions, when executed by a video processing apparatus, cause the video processing apparatus to perform a method for detecting a face in a video stream, the method comprising:

receiving, at the video processing apparatus, a sequence of input color images; and

for each input image, processing the sequence at the video processing apparatus, wherein the process includes,

calculating a greyscale image of the input color image,

creating a one-bit motion image based on the current greyscale image and a previous greyscale image,

calculating a normalized color image of the input color image,

calculating a motion color probability image by,

calculating a first color histogram of all the pixels in the normalized image,

calculating a second color histogram of the pixels in the normalized image corresponding to pixels in the motion image defined as being in motion,

calculating a ratio histogram by dividing the first color histogram by the second color histogram, and

setting pixel values in the motion color probability image to a value obtained by indexing the ratio histogram using values from the original normalized image,

providing at least the grayscale image and the motion color probability image to a face detector, and

executing face detection by determining a presence of a face based on first features in the grayscale image and second features in the motion color probability image.

16. The computer readable medium according to claim 15, wherein the first features include geometrical features or patterns.

17. The computer readable medium according to claim 15, wherein the second features include intensity information, the intensity information defining the presence or non-presence of assumed skin colors.

18. The computer readable medium according to claim 15, further comprising:

providing the motion image to the face detector, wherein the presence of a face is further based on third features in the motion image.

19. The computer readable medium according to claim 15, wherein the input color image is in the RGB color space, and wherein the normalized image is created by converting a copy of the input color image to the RG chromaticity space.

20. The method according to claim 15, wherein when the face detector detects a face in one of the input images, the calculating of the motion color probability image extracts the location of the detected face and calculates a second color histogram of the pixels representing the detected face.