REAL-TIME FACE DETECTION USING PIXEL PAIRS

• US 20140133758A1
• Filed: 11/09/2012
• Published: 05/15/2014
• Est. Priority Date: 11/09/2012
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Kienzle, Wolf
• Granted Patent: US 8,995,772 B2
• Time in Patent Office: Days
• US Class Current: 382/195
• CPC Class Codes: G06F 18/2148 characterised by the process organisation or structure, e.g. boosting cascade
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06V 10/446 using Haar-like filters, e.g. using integral image techniques
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

The subject disclosure is directed towards a face detection technology in which image data is classified as being a non-face image or a face image. Image data is processed into an image pyramid. Features, comprising pixel pairs of the image pyramid, are provided to stages of a cascading classifier to remove sub-window candidates that are classified as non-face sub-windows within each stage. The face detection technology continues with one or more subsequent stages to output a result as to whether the image contains a face.

Claims

1. In a computing environment, a method comprising, processing image data into an image pyramid, providing features of sub-window candidates to each stage of a cascading classifier to remove zero or more sub-window candidates that are classified as non-face sub-windows within that stage, in which the features comprise pixel pairs of the image pyramid, continuing with one or more subsequent stages, and outputting a result based upon whether at least one sub-window candidate remains.

2. The method of claim 1 further comprising, using luminance data to perform contrast normalization.

3. The method of claim 1 further comprising, using luminance data in at least one classifier to perform contrast normalization.

4. The method of claim 1 further comprising, padding a captured image in at least one direction into the image data that is processed into the image pyramid.

5. The method of claim 4 wherein padding the captured image comprises padding the captured image horizontally in a region based upon pixel data mirrored from an opposite region.

6. The method of claim 4 wherein padding the captured image comprises stretching the captured image vertically in a least one direction.

7. The method of claim 1 wherein continuing with the one or more subsequent stages ends at a last classifier, and wherein outputting the result comprises outputting a result indicating that a face was detected if at least one sub-window candidate remains.

8. The method of claim 1 wherein continuing with the one or more subsequent stages ends if no sub-window candidate remains, and wherein outputting the result comprises outputting a result indicating that no face was detected.

9. The method of claim 1 further comprising, using learned pixel-pair coordinates in at least one classifier.

10. The method of claim 1 further comprising, using at last one learned threshold value in at least one classifier.

11. The method of claim 1 further comprising, removing a sub-window candidate when combined differences of pixel pairs of that sub-window fail to achieve a global threshold value.

12. The method of claim 9 further comprising, learning the global threshold value.

13. The method of claim 1 wherein providing the features of the sub-window candidates to each stage of a cascading classifier comprises restricting the features provided to at least one stage to only features from certain pyramid levels.

16. The method of claim 1 further comprising, padding a captured image in at least one direction into the image data that is processed into the image pyramid.

14. A system comprising, at least one processor and memory, the memory including instructions, corresponding to a face detector, that are executed by the processor, the face detector configured to process image data into an image pyramid comprising representations of the image data at different resolutions, to extract features from the image pyramid comprising pixel pairs of the image pyramid, and to provide the features to a cascading classifier of the face detector, the cascading classifier configured to use the features to remove candidate sub-windows in a plurality of stages to detect whether the image data corresponds to a face or non-face image.

15. The system of claim 14 wherein at least one stage of the cascading classifier performs contrast normalization based upon a luminance value representative of a luminance of a sub-window versus a threshold luminance.

17. One or more computer-readable media having computer-executable instructions, which when executed perform steps, comprising,

(a) processing image data corresponding to the image into an image pyramid;

(b) obtaining a list of sub-window candidates;

(c) selecting a first classifier, of a plurality of classifiers comprising a cascading classifier, as a selected classifier;

(d) providing feature values corresponding to pixel-pairs of one or more sub-windows in the image pyramid to the selected classifier to remove any sub-window candidate from the list that does not achieve a threshold value associated with the selected classifier;

(e) determining whether a last classifier of the cascading classifier is the selected classifier, and if so, advancing to step (g);

(f) selecting a subsequent classifier of the cascading classifier as a selected classifier and returning to step (d);

(g) outputting a result indicative of whether at least one face was detected in the image data.

19. The one or more computer-readable media of claim 19 having further computer executable instructions comprising using luminance data associated with at least one sub-window for contrast normalization.

18. The one or more computer-readable media of claim 19 having further computer executable instructions comprising padding the captured image in at least one direction to provide the image data.

20. The one or more computer-readable media of claim 19 having further computer executable instructions comprising repeating steps (a) through (g) for each of a plurality of frames of images.