Method and apparatus for the automatic detection of facial features

• US 7,460,693 B2
• Filed: 09/27/2004
• Issued: 12/02/2008
• Est. Priority Date: 03/27/2002
• Status: Expired due to Fees

• Current Assignee: Seeing Machines LTD
• Original Assignee: Seeing Machines Pty Limited
• Inventors: Loy, Gareth, Thomsen, Colin
• Primary Examiner(s): Desire; Gregory M
• Application Number:
  US10/951,081
• Publication Number:
  US 20050041867A1
• Time in Patent Office: 1,527 Days
• Field of Search: 382/103, 382/117, 382/118, 382/190, 382/199
• US Class Current: 382/118
• CPC Class Codes: G06V 40/16 Human faces, e.g. facial parts, sketches or expressions

Abstract

A method of utilizing a computer system to automatically detect the location of a face within a series of images, the method comprising the steps of: detecting eye like regions within the series of images; utilizing the eye like regions to extract potential face regions within the series of images; enhancing the facial features in the extracted potential face regions; classifying the features; and verifying the face topology within the potential face regions.

Claims

1. A method of utilizing a computer system to automatically detect the location of a face within an input image, the method comprising the steps of:

(a) detecting a pair of eye locations within the input image;

(b) based on said pair of eye locations, extracting a potential face region within the input image;

(c) enhancing the extracted potential face region;

(d) detecting facial features within the enhanced extracted potential face region;

(e) classifying the facial features; and

(f) based on the classified facial features, verifying the face topology within the potential face region to determine the location of the face within the input image.

2. A method as claimed in claim 1 wherein said step

(a) further comprises the sub steps of:

(a1) identifying a plurality of eye candidates using a fast symmetry transform;

(a2) applying a Hough transform centered on each one of said plurality of eye candidates;

(a3) rating the quality of each of said Hough transformed plurality of eye candidates; and

(a4) selecting one pair of eye locations from said plurality of eye candidates based on one or more geometrical constraints.

3. A method as claimed in claim 1 wherein said step

(b) further comprises the sub steps of:

(b1) utilizing the location of said detected pair of eye locations to extract a region around said location to form said potential face region.

4. A method as claimed in claim 1 wherein said step

(c) further comprises the sub steps of:

(c1) utilizing local radial symmetry to construct a face mask of likely facial features within said potential face regions.

5. A method as claimed in claim 1 wherein said step

(c) further comprises the sub steps of:

(c1) reseating the intensity of the facial features within the potential face region.

6. A method as claimed in claim 1 wherein said step

(d) further comprises the sub steps of:

(d1) locating at least one of the mouth, nostrils or eyebrows of a face.

7. A method as claimed in claim 1 wherein said step of verifying the face topology includes using the spatial relationship between facial features in said images.

8. A method as claimed in claim 1 in which said at least one image is one of series of images.

9. A method as claimed in claim 1 wherein said step

(c) further includes the steps of:

(c1) constructing a gradient image of the face region; and

(c2) calculating a local radial symmetry of the gradient image of the face region.

10. A method as claimed in claim 1 wherein said step

(a) includes the steps of:

(a1) applying a dark radial symmetry transform to the input image so as to produce a first transformed image;

(a2) applying a Hough transform to the first transformed input image to produce a second transformed image.

11. A method as claimed in claim 10 wherein said step (a) further includes the steps of:

dividing the image into a number of regions;

determining the location of a maximum value of the dark radial symmetry transform in each region;

applying a Hough transform to the area centered on each maximum value so as to produce said second transformed image;

utilizing geometric constraints and relative image values in said second transformed image to determine potential eye locations within the input image.

12. A method as claimed in claim 1 wherein said step

(c) further includes the steps of:

(c1) constructing a gradient image of the face region; and

(c2) calculating a local radial symmetry of the gradient image of the face region.

13. A method of detecting at least one facial feature of a subject in an image, said method including the steps of:

(a) determining a pair of points, within the current image, representing the position of the subject'"'"'s eyes;

(b) selecting at least a portion of said current image, based on the position of said points within the current image;

(c) enhancing the selected portion of said current image;

(d) detecting one or more features within said enhanced image portion;

(e) classifying said one or more detected features as potential facial features; and

(f) determining whether at least one facial feature has been identified by comparing the relative positions of said one or more potential facial features within said enhanced image portion with at least one facial feature positioning condition.

14. A method as claimed in claim 13 wherein the step (a) includes the sub steps of:

identifying a plurality of candidate points using a fast symmetry transform;

applying a Hough transform centered on each one of said candidate points;

rating the quality of each of said plurality of candidate points, based on at least one of the following;

a fast symmetry transform, a Hough transform, a correlation measurement, or an intensity value centered on the candidate point; and

selecting one pair of points from said plurality of candidates points.

15. A method as claimed in claim 13 wherein said image is one image of series of images.

16. A method of detecting eyes within an input image, the method comprising the steps of:

(a) applying a dark radial symmetry transform to the input image so as to produce a first transformed image;

(b) dividing the image into a number of regions;

(c) determining a location of a maximum value of the dark radial symmetry transform in each region;

(d) applying a Hough transform to the area centered on each maximum value so as to produce a second transformed image; and

(e) analyzing geometric constraints and relative image values in said second transformed image to determine potential eye locations within the input image.