Method for Normalizing Displaceable Features of Objects in Images

• US 20110013804A1
• Filed: 07/16/2009
• Published: 01/20/2011
• Est. Priority Date: 07/16/2009
• Status: Active Grant

• Current Assignee: Fatih M. Porikli, Mohamed E. Hussein
• Original Assignee: Fatih M. Porikli, Mohamed E. Hussein
• Inventors: Hussein, Mohamed E., Porikli, Fatih M.
• Granted Patent: US 8,224,072 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06V 10/32 Normalisation of the pattern dimensions
• CPC Class Codes: G06V 10/462 Salient features, e.g. scale invariant feature transforms [SIFT]

Abstract

A method normalizes a feature of an object in an image. The feature of the object is extracted from a 2D or 3D image. The feature is displaceable within a displacement zone in the object, and wherein the feature has a location within the displacement zone. An associated description of the feature is determined. Then, the feature is displaced to a best location in the displacement zone to produce a normalized feature.

Claims

1. A method for normalizing a feature of an object in an image, comprising a processor for performing steps of the method, comprising:

extracting the feature of the object from the image, wherein the feature is displaceable within a displacement zone in the object, and wherein the feature has a location within the displacement zone;

determining an associated description of the feature; and

displacing the feature to a best location in the displacement zone to produce a normalized feature.

2. The method of claim 1, wherein the displaceable feature is used for detecting, recognizing, classifying or tracking the object.

3. The method of claim 1, wherein the object is rigid, deformable or articulated.

4. The method of claim 1, further comprising:

searching a set of training images iteratively for the best location, and wherein the searching uses a cost function.

5. The method of claim 1, further comprising:

applying a mapping function to directly displace the feature to the best location.

6. The method of claim 1, wherein the feature is F=(s, z₀, Z) with a size s, a location z₀, and a displacement zone Z.

7. The method of claim 1, wherein the displacement zone is defined by transformation parameters including similarity, affine, perspective motion transformations.

8. The method of claim 6, wherein the descriptor is Δ

_F(x, z₀), where x is the image from which the descriptor is extracted.

9. The method of claim 1, wherein the descriptor uses a histogram of oriented gradients of the feature.

10. The method of claim 4, wherein the feature is F=(s, z₀, Z), wherein a size of the feature is s, the location is z₀, and the displacement zone is Z, and wherein the descriptor is Δ

_F(x, z₀), and the cost function θ

(Δ

_F, z) measures a likelihood a descriptor at a location in a given training image matches the descriptor of the feature F at the location z₀.

11. The method of claim 10, wherein the cost function maximizes a likelihood score.

12. The method of claim 5, wherein the mapping function determines a displacement vector δ

=z*−

z₀, where z* is the best location and z₀ is the location of the feature.

13. The method of claim 10, wherein the cost function θ

measures a likelihood the object in the image is in a class of objects corresponding to the training images.

14. The method of claim 10, further comprising:

determining the descriptors Δ

_F(x, z₀) for each training image.

15. The method of claim 10, wherein the iterations terminate when a difference of successive estimates of the cost function is less than a predetermined threshold.

16. The method of claim 10, wherein the iterations terminate after a predetermined number of iterations.

17. The method of claim 1, wherein the image is three dimensional.

18. The method of claim 5, further comprising:

determining the mapping function from a set of training images.

19. The method of claim 1, further comprising:

normalizing the descriptor.

20. The method of claim 1, wherein the descriptor uses appearance based descriptors including edges, lines, pixel-wise filter responses, region templates, and statistical descriptors including region covariances, intensity histograms and co-occurrences.