Face detection device, face pose detection device, partial image extraction device, and methods for said devices

• US 7,123,754 B2
• Filed: 05/22/2002
• Issued: 10/17/2006
• Est. Priority Date: 05/22/2001
• Status: Active Grant

• Current Assignee: Panasonic Intellectual Property Corporation of America (Panasonic Holdings Corporation)
• Original Assignee: Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Inventors: Eshima, Toshirou, Baba, Naruatsu, Matsuo, Hideaki, Imagawa, Kazuyuki, Iwasa, Katsuhiro, Takata, Yuji
• Primary Examiner(s): Mancuso, Joseph
• Assistant Examiner(s): Akhavannik, Hadi
• Application Number:
  US10/154,404
• Publication Number:
  US 20020191818A1
• Time in Patent Office: 1,609 Days
• Field of Search: 382115-127, 382/199
• US Class Current: 382/118
• CPC Class Codes: G06T 7/74 involving reference images or patches
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation

Abstract

A face detection device includes a face learning dictionary, which holds learned information for identification between a facial image and a non-facial image. An image input unit inputs a subject image. An edge image extraction unit extracts an edge image from the subject image. A partial image extraction unit, based on the edge image, extracts partial images that are candidates to contain facial images from the subject image. A face/non-face identification unit references the learning dictionary to identify whether or not each extracted partial image contains a facial image. Face detection of high precision, which reflects learned results, is performed.

Claims

1. A face detection device comprising:

a face learning dictionary for holding learned information for identification between a facial image and a non-facial image;

an image input unit operable to input a subject image;

an edge image extraction unit operable to extract an edge image from said subject image;

a partial image extraction unit operable to extract, based on said edge image, partial images that are candidates to contain facial images from said subject image;

a face/non-face identification unit operable to reference said learning dictionary and identifies whether or not each extracted partial image contains a facial image,

wherein the partial image extraction unit is configured to determine a contour that joins upper end parts of the edge image and further configured to extract partial images from the subject image based on peaks and valleys of the contour.

2. The face detection device as set forth in claim 1, wherein:

said image input unit inputs a subject image comprising a moving image; and

said edge image extraction unit extracts an edge image by time subtraction of frames of said moving image.

3. The face detection device as set forth in claim 1, wherein:

said image input unit inputs a subject image comprising a still image; and

said edge image extraction unit extracts an edge image by applying a differential filter to said still image.

4. The face detection device as set forth in claim 1, wherein said face learning dictionary contains both learned information based on facial images and learned information based on non-facial images.

5. The face detection device as set forth in claim 1, wherein said face/non-face identification unit calculates a predefined evaluation function value and identifies between a face and a non-face based on said evaluation function value.

6. The face detection device as set forth in claim 5, wherein a characteristic vector, having luminance information on a partial image as components, is used in said evaluation function.

7. The face detection device as set forth in claim 5, wherein a characteristic vector, having angles of edge normal direction vectors, which were derived from a partial image, as components, is used in said evaluation function.

8. The face detection device as set forth in claim 5, wherein a characteristic vector is expressed by a histogram, which was derived from a partial image.

22. The face detection device as set forth in claim 1, wherein the partial image extraction unit is further configured to determine a first contour of said edge image, to filter said first contour to obtain a second contour having peaks and valleys, and to extract, based on said peaks and valleys, partial images that are candidates to contain facial images from said subject image.

9. A face pose detection device comprising:

a face learning dictionary, which holds learned information for identification between a facial image and a non-facial image;

an image input unit operable to input a subject image;

an edge image extraction unit operable to extract an edge image from said subject image;

a partial image extraction unit operable to extract, based on said edge image, partial images that are candidates to contain facial images from said subject image;

a face pose learning dictionary, which holds learned information on a plurality of predefined face poses according to each face pose;

an evaluation function value calculation unit operable to calculate evaluation function values for all predefined face poses based on said learned information on said face poses and said extracted partial images; and

a face pose calculation unit operable to determine said face pose based on said plurality of evaluation function values that have been calculated, including a face pose that is intermediate among said predefined face poses,

wherein the partial image extraction unit is configured to determine a contour that joins upper end parts of the edge image and further configured to extract partial images from the subject image based on peaks and valleys of the contour.

10. The face pose detection device as set forth in claim 9, wherein said predefined plurality of face poses are said face poses of four directions of front, back, right, and left and said face pose calculation unit determines a direction of a synthetic vector of said evaluation function values for these four directions as said face pose to be determined.

11. A partial image extraction device comprising:

an image input unit operable to input a subject image;

an edge image extraction unit operable to extract an edge image from said subject image;

a partial image extraction operable to extract, based on said edge image, partial images that are considered to contain facial images from said subject image; and

said partial image extraction unit is configured to determine a contour that joins upper end parts of the edge image and further configured to extract partial images from said subject image based on peaks and valleys of the contour.

12. A face detection method comprising:

inputting a subject image;

extracting an edge image is extracted from said subject image;

based on said edge image, extracting partial images from said subject image that are candidates to contain facial images;

referencing a face learning dictionary that holds learned information for identification between a facial image and a non-facial image to determine whether or not each extracted partial image contains a facial image;

determining a contour that joins upper end parts of said edge image; and

extracting partial images from said subject image based on peaks and valleys of said contour.

13. The face detection method as set forth in claim 12, wherein said subject image is a moving image and extracting an edge image by time subtraction of frames of said moving image.

14. The face detection method as set forth in claim 12, wherein said subject image is a still image, and applying a differential filter to said still image to extract an edge image.

15. The face detection method as set forth in claim 12, wherein said face learning dictionary contains both learned information based on facial images and learned information based on non-facial images.

16. The face detection method as set forth in claim 12, further comprising:

calculating a predefined evaluation function value; and

evaluating said evaluation function value to determine identification between a face and a non-face.

17. The face detection method as set forth in claim 16, wherein a characteristic vector, having luminance information on a partial image as components, is used in said evaluation function.

18. The face detection method as set forth in claim 16, wherein a characteristic vector, having angles of edge normal direction vectors, which were derived from a partial image, as components, is used in the evaluation function.

19. The face detection method as set forth in claim 16, wherein a characteristic vector is expressed by a histogram, which was derived from a partial image.

23. The face detection method as set forth in claim 12, further comprising the steps of:

applying a filter to said edge image to determine a first contour having peaks and valleys;

applying a filter to said first contour to obtain a second contour having peaks and valleys; and

extracting partial images from said subject image based on said peaks and valleys of said second contour.

20. A face pose detection method comprising the steps of:

inputting a subject image;

extracting an edge image from said subject image;

determining a contour that joins upper end parts of said edge image;

extracting partial images from said edge image that are candidates to contain facial images based on peaks and valleys of said contour;

calculating an evaluation function value, which includes referencing a face pose learning dictionary, which holds learned information on a plurality of predefined face poses according to each face pose;

the step of calculating includes calculating said evaluation function values for all predefined face poses based on said learned information on said face poses and said extracted partial images; and

calculating a face pose, based on said plurality of evaluation function values that have been calculated, said face pose, including determining a face pose that is intermediate among said predefined face poses.

21. The face pose detection method as set forth in claim 20, wherein said predefined plurality of face poses are face poses of the four directions of front, back, right, and left and determining a direction of a synthetic vector of the evaluation function values for these four directions as the determined face pose.