Representing Object Shapes Using Radial Basis Function Support Vector Machine Classification

• US 20120207384A1
• Filed: 02/14/2011
• Published: 08/16/2012
• Est. Priority Date: 02/14/2011
• Status: Active Grant

• Current Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Inventors: Porikli, Fatih, Nguyen, Hien
• Granted Patent: US 8,718,380 B2
• Time in Patent Office: Days
• US Class Current: 382/154
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: G06V 10/46 Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features colour feature extraction G06V10/56
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 20/64 Three-dimensional objects

Abstract

A shape of an object is represented by a set of points inside and outside the shape. A decision function is learned from the set of points an object. Feature points in the set of points are selected using the decision function, or a gradient of the decision function, and then a local descriptor is determined for each feature point.

Claims

1. A method for representing a shape of an object represented by a set of points, wherein the set of points include points inside, outside and a boundary of the shape, the method comprising the steps of

learning a decision function from the set of points;

selecting feature points in the set of points using the decision function; and

determining a local descriptor for each feature point, wherein the steps are performed in a processor.

2. The method of claim 1, further comprising:

determining a gradient of the decision function, and wherein the selecting uses the gradient of the decision function.

3. The method of claim 1, wherein the points are pixels in a 2D image of the object.

4. The method of claim 1, wherein the points are samples on a 3D shape of the object.

5. The method of claim 1, wherein the decision function is obtained by training a discriminative classifier using the set of points.

6. The method of claim 1, wherein the decision function is a radial basis function (RBF) support vector machine (SVM).

7. The method of claim 6, wherein the decision function is a hypersurface defined by support vectors of the SVM that separates the set of points inside and outside the shape.

8. The method of claim 1, wherein the decision function is positive for the points inside shape, zero for the points on a boundary of the shape, and negative for the points outside the shape.

9. The method of claim 1, wherein the learning uses subsets of randomly selected points.

10. The method of claim 1, further comprising;

normalizing the shape.

11. The method of claim 10, further comprising;

determining a scale parameter by a mean pair-wise distances of the set of points; and

,

dividing the original point coordinates using the scale parameter.

13. The method of claim 1, wherein the feature points are based on gradients of the decision function for the set of points.

14. The method of claim 1, wherein the feature points are based on a curvature of the decision function curvature for the set of points.

15. The method of claim 1, wherein the feature points are based on gradients of entropy of the decision function for the set of points.

16. The method of claim 1, wherein the local descriptors are histograms of oriented gradients of decision function around each point.

17. The method of claim 1, wherein the local descriptors are spin images of the decision function.

18. The method of claim 1, wherein the local descriptors are co-centric ring signatures of the decision function.

19. The method of claim 1, further comprising:

classifying unknown shapes according to the feature points and their local descriptors.

20. The method of claim 19, further comprising:

constructing an ordered list of the feature points of an object by ordering the feature points according to angles from the center of the object in a circular fashion.

21. The method of claim 20, further comprising:

selecting the feature point that has a highest gradient magnitude;

setting the selected feature point as a first element of the list; and

assigning the remaining feature points in a circular courterclock-wise in the ordered list.

22. The method of claim 20, further comprising:

determining a match distance for the ordered lists of two 2D objects.

23. The method of claim 22, further comprising:

determining a cost function that represents an overall distance for corresponding pairing of the descriptors using a descriptor distance;

minimizing the cost function by dynamic programming to obtain a mapping function; and

determining a match distance using the mapping function.

24. The method of claim 6, wherein kernels of the RBF provide affine invariant features.

25. The method of claim 6, wherein the decision function is analytic, and in a form of a weighted sum of responses of the kernels.

26. The method of claim 1, wherein the local descriptors are invariant to translation and rotation of the shape.