Invariant visual scene and object recognition

• US 8,036,468 B2
• Filed: 12/24/2007
• Issued: 10/11/2011
• Est. Priority Date: 12/24/2007
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Chrysanthakopoulos, Georgios
• Primary Examiner(s): CHAWAN, SHEELA C
• Application Number:
  US11/963,858
• Publication Number:
  US 20090161968A1
• Time in Patent Office: 1,387 Days
• Field of Search: 382/181, 382/103, 382/190, 382/173, 382/232, 382/243, 382/224, 382/228, 382/209, 382/217, 382/218, 382/100, 382/195, 382/199, 382/276, 382/293, 382/296, 382/107, 382/192, 382/286, 382/289, 382/128, 382/186, 382/108, 382/254, 382/275, 382/168, 382/171, 382/278, 382/294, 382/225, 358/400, 358/443, 358/448, 358/462, 367/88, 356/27, 356/28, 257/E21.53, 257/E21.001, 257/E21.529
• US Class Current: 382/225
• CPC Class Codes: G06V 10/758 Involving statistics of pixels or of feature values, e.g. histogram matching

Abstract

A method for performing image recognition is disclosed. The method includes obtaining a collection of pixels and grouping at least some of the pixels into a set of cluster features based on gradient magnitude. For each cluster feature in the set, statistical variables are generated. The statistical variables represent a collective property of the pixels in the cluster feature. The statistical variables are utilized as a basis for comparing the collection of pixels to a different collection of pixels.

Claims

1. A computer-implemented image processing method, the method comprising:

generating a collection of gradient data associated with a first image;

assigning pixels within the first image into a set of cluster features based on gradient magnitude as reflected in the collection of gradient data; and

storing the set of cluster features.

2. The method of claim 1, wherein assigning comprises assigning pixels such that the pixels in a given cluster feature have a same or substantially similar gradient magnitude.

3. The method of claim 1, further comprising:

generating a second collection of gradient data associated with a second image;

assigning pixels within the second image into a second set of cluster features based on gradient magnitude as reflected in the second collection of gradient data; and

storing the second set of cluster features.

4. The method of claim 3, further comprising comparing the first image to the second image by evaluating the first set of cluster features relative to the second set of cluster features.

5. The method of claim 4, wherein evaluating the first set of cluster features relative to the second set of cluster features comprises evaluating a first statistical variable to a second statistical variable, wherein the first statistical variable is indicative of a property of pixels included in a first cluster feature that is part of the first set of cluster features, and wherein the second statistical variable is indicative of a property of pixels included in a first cluster feature that is part of the second set of cluster features.

6. The method of claim 5, wherein the first statistical variable is indicative of a property that is calculated by determining an average across all pixels in the first cluster feature.

7. The method of claim 3, wherein comparing the first image to the second image comprises aligning cluster features in the first set with cluster features in the second set based on similarity of the cluster features.

8. The method of claim 1, further comprising, for each cluster feature in the set, calculating and storing a set of statistical variables that are indicative of properties of pixels included in the cluster feature.

9. The method of claim 8, wherein the set of statistical variables includes an area variable indicative of the number of pixels in the cluster feature.

10. The method of claim 8, wherein the set of statistical variables includes a variable indicative of a relative position of pixels in the cluster feature.

11. The method of claim 8, wherein the set of statistical variables includes an average gradient angle variable that is collectively indicative of pixels within the cluster feature.

12. The method of claim 8, wherein the set of statistical variables includes a skew variable.

13. The method of claim 8, wherein the set of statistical variables includes a standard deviation variable.

14. A computer-implemented method for performing image recognition, the method comprising:

obtaining a collection of pixels;

grouping at least some of the pixels in the collection into a set of cluster features based on gradient magnitude;

for each cluster feature in the set, generating statistical variables that represent a collective property of the pixels in the cluster feature; and

utilizing the statistical variables as a basis for comparing the collection of pixels to a different collection of pixels.

15. The method of claim 14, wherein utilizing the statistical variables comprises comparing cluster features in the set to cluster features in a different set associated with the different collection of pixels.

16. The method of claim 14, wherein generating statistical variables comprises generating an average color of pixels included in the cluster feature.

17. The method of claim 14, wherein generating statistical variables comprises generating an average position of pixels included in the cluster feature.

18. The method of claim 14, wherein generating statistical variables comprises generating a number of pixels included in the cluster feature.

19. A computer-implemented method for performing image recognition, the method comprising determining the relative similarity of a first image to a second image, wherein determining comprises comparing cluster features associated with the first image to cluster features associated with the second image, and wherein the cluster features associated with the first and second images are derived based on pixels that have been grouped together based on gradient magnitude.

20. The method of claim 19, wherein determining further comprises comparing the number of cluster features associated with the first image to the number of cluster features associated with the second image.