LOGO OR IMAGE RECOGNITION

• US 20120263385A1
• Filed: 04/15/2011
• Published: 10/18/2012
• Est. Priority Date: 04/15/2011
• Status: Active Grant

• Current Assignee: Verizon Patent and Licensing Incorporated (Verizon Communications Inc.)
• Original Assignee: Yahoo! Inc. (Apollo Global Management, Inc.)
• Inventors: van Zwol, Roelof, Pueyo, Luis Garcia, Romberg, Stefan
• Granted Patent: US 8,634,654 B2
• Time in Patent Office: Days
• US Class Current: 382/201
• CPC Class Codes: G06V 10/757 Matching configurations of points or features

Abstract

Subject matter disclosed herein relates to electronic image object or logo recognition.

Claims

1. A method for identifying a logo or object in a particular image, the method comprising:

quantizing image descriptors for said particular image;

searching for matches between selected groups of the quantized image descriptors and groups in a group index of a collection of logos or objects; and

if matches occur, searching for further matches between larger groups of said quantized image descriptors and polygons in a polygon index of said collection of logos or objects, wherein said larger groups include the matched selected groups.

2. The method of claim 1, wherein said polygons comprise triangles, and wherein said searching for matches further comprises:

quantizing said triangles.

3. The method of claim 2, wherein the quantized triangles comprise signatures that include quantized components that are based, at least in part, on positions of said image descriptors relative to one another.

4. The method of claim 2, wherein said triangles are approximately scale invariant.

5. The method of claim 1, wherein said groups of said quantized image descriptors comprise pairs of said quantized image descriptors and said larger groups of said quantized image descriptors comprise triplets of said quantized image descriptors.

6. The method of claim 5, wherein said triplets comprise three points of triangles in said particular image.

7. The method of claim 1, wherein said image descriptors comprise stable points of an object in said particular image, wherein relative positions of said stable points are substantially constant with respect to varying perspectives of said object;

8. The method of claim 1, wherein said identifying a logo or object comprises identifying said logo or object with a false positive rate that is approximately or less than 1/200 that of a false negative rate.

9. The method of claim 1, further comprising:

using a pseudo random process to select said selected groups.

10. The method of claim 9, wherein said pseudo random process comprises a Monte Carlo process.

11. An apparatus comprising:

a computing platform to perform a cascaded process, wherein said cascaded process comprises comparing points extracted from a sample image to components of a primary index, and querying a secondary index if said points correspond to components of said primary index.

12. The apparatus of claim 11, wherein said primary index comprises quantized pairs of points extracted from two or more training images.

13. The apparatus of claim 12, wherein individual said two or more training images include a logo or object and said sample image includes said logo or object if said querying said secondary index produces a match among components in said secondary index and one or more quantized combinations of said points.

14. The apparatus of claim 12, wherein said secondary index comprises combinations of said quantized pairs of points.

15. An apparatus comprising:

a special purpose computing system, said special purpose computing system to;

receive electronic signals representing an image;

transform electronic signals to encode a geometric structure of local features of an object in said image; and

identify said object based, at least in part, on logo-classes generated from multiple training images.

16. The apparatus of claim 15, wherein said multiple training images include a collection of logos or objects.

17. The apparatus of claim 15, wherein said geometric structure comprises hessian-affine features.

18. The apparatus of claim 15, said special purpose computing system further to select advertisements based, at least in part, on the identifying said object.

19. The apparatus of claim 18, wherein the selected advertisements include the identified object.

20. The apparatus of claim 19, wherein said identified object comprises a logo.