Methods and system for recognizing people with model-based face detection

• US 6,999,606 B1
• Filed: 10/04/1999
• Issued: 02/14/2006
• Est. Priority Date: 10/05/1998
• Status: Expired due to Term

• Current Assignee: BioID AG
• Original Assignee: HumanScan GmbH
• Inventors: Frischholz, Robert
• Primary Examiner(s): Bali, Vikkram
• Application Number:
  US09/806,785
• Time in Patent Office: 2,325 Days
• Field of Search: 382115-118, 382/140, 235/380, 235/382, 235/382.5, 358/296, 340/5.53, 340/5.83, 902 2- 6
• US Class Current: 382/118
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation

Abstract

The invention relates to a face finding method and system in a binarized image by comparing the dot group of the binarized image with the dot group of a face model, wherein the dot groups of the binarized image and of the face model are compared on the basis of the Hausdorff spacing between the dots of the dot groups and a face is recognized in the binarized image when a measure derived from the Hausdorff spacing fails to reach a limit value.

Claims

1. A method of finding a face in a binarized image by comparing the dot group of the binarized image with the dot group of a face model, the face model being subjected to multiple two-dimensional transforms in order to locate the face model in the binarized image, and the dot groups of the binarized image and the face model being compared on the basis of the Hausdorff spacing between the dots of the dot groups and a position of a face in a binarized image is found when a measure derived from the Hausdorff spacing fails to reach a limit value.

2. The method as claimed in claim 1, wherein the binarized image is derived from the original image by means of edge extraction.

3. The method as claimed in claim 2, wherein the binarized image first is compared on a small scale with a face model of corresponding small size, the area of the binarized image in which a face was found is enlarged and compared once again with a face model of corresponding larger size, the enlarging and comparing of the binarized image area and face model are repeated, as the case may be, until the face in the binarized image was localized with sufficient accuracy.

4. The method as claimed in claim 3, wherein different face models with different resolutions are used depending on the size of the binarized image.

5. The method as claimed in claim 4, wherein the edge extraction for deriving the binarized image from the original image is carried out with different resolutions depending on the size of the binarized image.

6. The method as claimed in claim 5, wherein the face model and/or the binarized image are rotated and different steps of the rotation are used depending on the size of the binarized image.

7. The method as claimed in claim 4, wherein the face model and/or the binarized image are rotated and different steps of the rotation are used depending on the size of the binarized image.

8. The method as claimed in claim 3, wherein the edge extraction for deriving the binarized image from the original image is carried out with different resolutions depending on the size of the binarized image.

9. The method as claimed in claim 8, wherein the face model and/or the binarized image are rotated and different steps of the rotation are used depending on the size of the binarized image.

10. The method as claimed in claim 3, wherein the face model and/or the binarized image are rotated and different steps of the rotation are used depending on the size of the binarized image.

11. The method as claimed in claim 1, wherein the binarized image first is compared on a small scale with a face model of corresponding small size, the area of the binarized image in which a face was found is enlarged and compared once again with a face model of corresponding larger size, the enlarging and comparing of the binarized image area and face model are repeated, as the case may be, until the face in the binarized image was localized with sufficient accuracy.

12. The method as claimed in claim 11, wherein different face models with different resolutions are used depending on the size of the binarized image.

13. The method as claimed in claim 12, wherein the binarized image is derived from the original image by means of edge extraction and, the edge extraction for deriving the binarized image from the original image is carried out with different resolutions depending on the size of the binarized image.

14. The method as claimed in claim 13, wherein the face model and/or the binarized image are rotated and different steps of the rotation are used depending on the size of the binarized image.

15. The method as claimed in claim 12, wherein the face model and/or the binarized image are rotated and different steps of the rotation are used depending on the size of the binarized image.

16. The method as claimed in claim 11, wherein the binarized image is derived from the original image by means of edge extraction and, the edge extraction for deriving the binarized image from the original image is carried out with different resolutions depending on the size of the binarized image.

17. The method as claimed in claim 16, wherein the face model and/or the binarized image are rotated and different steps of the rotation are used depending on the size of the binarized image.

18. The method as claimed in claim 11, wherein the face model and/or the binarized image are rotated and different steps of the rotation are used depending on the size of the binarized image.

19. The method as claimed in claim 1, wherein in the Hausdorff measure is determined on the basis of the average value of a certain percentage of the smallest minimum Hausdorff spacings, the percentage being between 0% and 100%.

20. A system for implementing the method as claimed in claim 19, comprising a computing device for calculating the Hausdorff spacing and the Hausdorff measure on the basis of the dots of the binarized image and the face model.

21. A system for implementing the method as claimed in claim 1, comprising a computing device for calculating the Hausdorff spacing and the Hausdorff measure on the basis of the dots of the binarized image and the face model.