Method of recognizing faces using range images

• US 6,137,896 A
• Filed: 10/07/1997
• Issued: 10/24/2000
• Est. Priority Date: 10/07/1997
• Status: Expired due to Fees

• Current Assignee: National Research Council Canada
• Original Assignee: National Research Council Canada
• Inventors: Chang, Shoude, Rioux, Marc
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Werner, Brian P.
• Application Number:
  US08/944,954
• Time in Patent Office: 1,113 Days
• Field of Search: 382/118, 382/124-127, 382/191, 382/210, 382/211, 382/280, 382/283, 382/278, 340/825.34, 356/71, 356/394, 349/17, 324/76.33
• US Class Current: 382/118
• CPC Class Codes: G06V 10/88 Image or video recognition using optical means, e.g. reference filters, holographic masks, frequency domain filters or spatial domain filters
• CPC Class Codes: G06V 20/647 by matching two-dimensional images to three-dimensional objects
• CPC Class Codes: G06V 40/16 Human faces, e.g. facial parts, sketches or expressions

Abstract

Faces are one of the most commonly recognized features among people. According to the invention a system for recognizing intensity images of faces is provided. Intensity images of a face are captured and normalized. The intensity images are transformed, filtered and then transformed again. The filter is based on a three-dimensional image of a face against which the intensity image is being compared. The result is a form of correlation that is substantially illumination independent.

Claims

1. A method of comparing a three-dimensional facial image with a two-dimensional intensity image of a face comprising the steps of:

providing a filter having three orthogonal components derived from a plurality of normal vectors derived from points defining the surface of the three-dimensional facial image;

performing a correlation by the steps of providing a transform of at least one of the two-dimensional intensity image other than a transform of the intensity image into a three-dimensional image and filtering the provided transform using the filter derived from three-dimensional facial image to obtain correlation results which are partially independent of the angle from which a face within the two-dimensional intensity image is illuminated; and

,

when the correlation results are within a first range, providing a signal indicative of a match.

2. A method of comparing a three-dimensional facial image with a two-dimensional intensity image of a face as defined in claim 1 wherein the correlation results are substantially independent of angle of illumination of a face within the two dimensional intensity image.

3. A method of comparing a three-dimensional facial image with a two-dimensional intensity image of a face as defined in claim 1 wherein the step of performing a correlation comprises the steps of transforming the two-dimensional intensity image before filtering the provided transform, and transforming the filtered transformed two-dimensional intensity image to produce the correlation result.

4. A method of comparing a three-dimensional facial image with a two-dimensional intensity image of a face as defined in claim 3 wherein the filter comprises a phase-only filter determined based on the three-dimensional image.

5. A method of comparing a three-dimensional facial image with a two-dimensional intensity image of a face as defined in claim 4 wherein the two-dimensional intensity image is transformed according to one of a Fourier transform, binary transform, and non-linear joint transform and the filtered transformed two-dimensional intensity image is transformed according to an inverse of said transform.

6. A method of comparing a three-dimensional facial image with a two-dimensional intensity image of a face as defined in claim 5 comprising the step of repeating the step of performing a correlation for each of a plurality of different three-dimensional images.

7. A method of comparing a three-dimensional facial image with a two-dimensional intensity image of a face as defined in claim 1 wherein the two-dimensional intensity image is transformed according to one of a Fourier transform, binary transform, and non-linear joint transform.

8. A method of comparing a three-dimensional facial image with a two-dimensional intensity image of a face as defined in claim 1 wherein the two-dimensional intensity image comprises pixels, the method comprising the step of normalising pixels within the two-dimensional intensity image.

9. A method of comparing a three-dimensional facial image with a two-dimensional intensity image of a face as defined in claim 1 wherein the correlation is performed substantially according to the following function:

space="preserve" listing-type="equation">|c|.sup.2 ≅

ι

.sub.x.sup.2 |p.sub.x (x,y)Xƒ

.sup.i.sub.x |.sup.2 +ι

.sub.y.sup.2 |p.sub.y (x,y)Xƒ

.sup.i.sub.y |.sup.2 +ι

.sub.z.sup.2 |p.sub.z (x,yXƒ

.sup.i.sub.z |.sup.2.

10. A system for performing facial recognition comprising:

means for providing a two-dimensional intensity image of a face;

means for transforming the intensity image in a predetermined fashion other than a transform of the intensity image into a three-dimensional image;

means for filtering the transformed intensity image with a filter having three orthogonal components derived from a plurality of normal vectors derived from points defining the surface of a face within a known three-dimensional image for producing a filtered transform image that is substantially independent of an angle of illumination of a face within the two-dimensional intensity image;

means for transforming the filtered transformed intensity image to produce a result indicative of a correlation between he provided intensity image of a face and the known three-dimensional image.

11. A system for performing facial recognition according to claim 10 wherein the means for transforming the intensity image comprises means for one of Fourier transform, binary transform, and non-linear joint transform.

12. A system for performing facial recognition according to claim 10 wherein the means for transforming the filtered transformed intensity image comprises means for performing substantially the inverse transform to that performed by the means for transforming the image.

13. A system for performing facial recognition according to claim 10 wherein the means for transforming the intensity image comprises Fourier transform means and wherein the means for transforming the filtered transformed intensity image comprises inverse Fourier transform means.

14. A system for performing facial recognition according to claim 13 wherein the means for transforming the intensity image and the means for transforming the filtered transformed intensity image comprise lenses for performing the transforms on an optical signal.

15. A system for performing facial recognition according to claim 10 comprising means for determining a phase only filter in dependence upon a known three-dimensional image of a face for use by the means for filtering the transformed image.

16. A system for performing facial recognition as defined in claim 10 wherein the means for providing a two-dimensional intensity image of a face comprises image capture means for capturing a first intensity image and for providing a first signal in dependence upon the captured image;

and,

wherein the means for transforming the intensity image, the means for filtering the transformed intensity image, and the means for transforming the filtered transformed intensity image comprise at least a processor for receiving the first signal and for applying a transform to the first signal to provide a transformed first signal, for filtering the transformed first signal, the filter determined in dependence upon a three-dimensional image of a face and for applying an inverse of the transform to the filtered transformed first signal.

17. A system for performing facial recognition as defined in claim 10 comprising

a light source for providing a beam of light;

a filter for filtering the beam of light, the filter for receiving a first signal and for filtering in dependence thereon, and

wherein the means for providing a two-dimensional intensity image of a face comprises

image capture means for capturing a first intensity image and for providing the first signal in dependence thereon;

wherein the means for transforming the intensity image in a predetermined fashion comprises a lens for receiving the filtered beam of light and for transforming the beam of light in a predetermined fashion;

wherein the means for filtering the transformed intensity image in dependence upon a known three-dimensional image comprises a second filter for filtering the beam of light,

the second filter for receiving a second signal in dependence upon a known three-dimensional image and for filtering in dependence thereon; and

,

wherein the means for transforming the filtered transformed intensity image to produce a result indicative of a correlation between the provided intensity image of a face and the known three-dimensional image comprises a lens for receiving the filtered transformed beam of light and for transforming the beam of light in substantially an inverse fashion to the predetermined fashion.

18. A system for performing facial recognition as defined in claim 17 wherein the second filter comprises a spatial light modulator.

19. A system for performing facial recognition as defined in claim 10 wherein the means for providing a two-dimensional intensity image of a face consists of a single two dimensional intensity image capture means.

20. A method of comparing a three-dimensional facial image with a two-dimensional intensity image of a face comprising the steps of:

providing information derived from the three-dimensional facial image;

extracting three orthogonal images from the three-dimensional image, the orthogonal images based on orthogonal components of vectors normal to the surface of the three-dimensional image;

performing a correlation of a transform of the two-dimensional intensity image and extracted orthogonal images to obtain correlation results which are partially independent of the angle from which a face within the two-dimensional intensity image is illuminated; and

,

when the correlation results are within a first range, providing a signal indicative of a match.

21. A method of comparing a three-dimensional facial image with a two-dimensional intensity image of a face as defined in claim 20 wherein the three orthogonal images are used to define a phase-only filter according to