IMAGE CONCEALING VIA EFFICIENT FEATURE SELECTION

• US 20140321758A1
• Filed: 07/03/2014
• Published: 10/30/2014
• Est. Priority Date: 01/09/2012
• Status: Active Grant

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Liu, Xiaoming, Tu, Peter Henry, Xue, Ya
• Granted Patent: US 9,082,043 B2
• Time in Patent Office: Days
• US Class Current: 382/215
• CPC Class Codes: G06F 18/2113 by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
• CPC Class Codes: G06F 18/2178 based on feedback of a supervisor
• CPC Class Codes: G06F 18/2323 based on graph theory, e.g. minimum spanning trees [MST] or graph cuts
• CPC Class Codes: G06V 10/754 involving a deformation of the sample pattern or of the reference pattern; Elastic matching
• CPC Class Codes: G06V 10/7635 based on graphs, e.g. graph cuts or spectral clustering
• CPC Class Codes: G06V 10/771 Feature selection, e.g. selecting representative features from a multi-dimensional feature space
• CPC Class Codes: G06V 10/7784 based on feedback from supervisors
• CPC Class Codes: G06V 40/171 Local features and components; Facial parts eye characteristics G06V40/18; Occluding parts, e.g. glasses; Geometrical relationships

Abstract

A novel technique for unsupervised feature selection is disclosed. The disclosed methods include automatically selecting a subset of a feature of an image. Additionally, the selection of the subset of features may be incorporated with a congealing algorithm, such as a least-square-based congealing algorithm. By selecting a subset of the feature representation of an image, redundant and/or irrelevant features may be reduced or removed, and the efficiency and accuracy of least-square-based congealing may be improved.

Claims

1. A method, comprising:

integrating an unsupervised feature selection algorithm with a least-square based congealing algorithm;

selecting a subset of features from an initial feature representation with the unsupervised feature selection algorithm; and

executing the least-square based congealing algorithm to estimate warping parameters for a plurality of images in an ensemble using the subset of features.

2. The method of claim 1, wherein the subset of features comprises less than approximately 3% of the initial feature representation.

3. The method of claim 1, wherein the subset of features comprises a first representative feature from a first subset of the initial feature representation and a second representative feature from a second subset of the initial feature representation.

4. The method of claim 3, comprising dividing the initial feature representation into the first subset and the second subset automatically.

5. The method of claim 3, comprising selecting the first representative feature and the second representative feature based on a feature similarity measure.

6. The method of claim 1, wherein selecting a subset of features from an initial feature representation with the unsupervised feature selection algorithm comprises using a filter method to select the subset of features from the initial feature representation.

7. The method of claim 1, wherein using a filter method to select the subset of features from the initial feature representation comprises removing irrelevant features and/or redundant features of the initial feature representation from the subset of features.

8. The method of claim 1, wherein executing the least-square based congealing algorithm to estimate warping parameters for the plurality of images in the ensemble using the subset of features comprises estimating initial warping parameters, computing warping parameter updates, and updating the initial warping parameters to calculate current warping parameters.

9. A method, comprising:

incorporating an unsupervised feature selection algorithm into an image congealing algorithm;

executing the unsupervised feature selection algorithm, comprising;

constructing a graph having features of an image as vertices;

determining a connectivity between the vertices;

partitioning the graph into two or more subsets of features; and

selecting representative features from each subset of features; and

executing the image congealing algorithm to estimate warping parameters for the image using the representative features.

10. The method of claim 9, wherein the image congealing algorithm is a least-square based congealing algorithm.

11. The method of claim 9, comprising determining the connectivity between the vertices using a feature similarity measure.

12. The method of claim 11, wherein the feature similarity measure comprises a maximum information compression index, a heuristic algorithm, a spectral clustering algorithm, or any combination thereof.

13. The method of claim 9, comprising partitioning the graph into two or more subsets of features using a power iteration clustering algorithm, wherein the power iteration clustering algorithm comprises a perturbation enhancement.

14. The method of claim 9, comprising partitioning the graph into two or more subsets of features using a Dirichlet process mixture model.

15. The method of claim 9, comprising executing multiple iterations of the image congealing algorithm and executing the unsupervised feature selection algorithm before each iteration of the image congealing algorithm.

16. The method of claim 9, wherein a number of the two or more subsets of features is determined automatically.

17. The method of claim 9, wherein a number of the two or more subsets of features is predetermined, and a number of representative features from each subset of features is predetermined.

18. A method, comprising:

minimizing a cost function of an image congealing process, comprising;

selecting a subgroup of features for each of a plurality of original feature representations with an unsupervised feature selection algorithm, wherein each of the plurality of original feature representations corresponds to a respective image of a plurality of images in an ensemble; and

estimating warping parameters for each of the plurality of images using the subgroups of features in a least-square-based congealing algorithm.

19. The method of claim 18, wherein the unsupervised feature selection algorithm comprises:

generating a graph for each image, wherein vertices of the graph comprise the features of the original feature representation;

determining connectivity between the vertices with a maximum information compression index;

dividing the graph into subsets using a power iteration clustering method; and

selecting a representative feature from each subset to determine the subgroup of features.

20. The method of claim 18, wherein a number of selected features in each subgroup is less than approximately 3% of a total number of features of the respective original feature representation.