METHOD FOR IMPROVING CLASSIFICATION RESULTS OF A CLASSIFIER

• US 20140219553A1
• Filed: 03/02/2012
• Published: 08/07/2014
• Est. Priority Date: 03/04/2011
• Status: Active Grant

• Current Assignee: LBT Innovations Ltd.
• Original Assignee: Anton John Van Den Hengel
• Inventors: van den Hengel, Anton John, Hill, Rhys Ernst, Ward, Benjamin William
• Granted Patent: US 10,037,480 B2
• Time in Patent Office: Days
• US Class Current: 382/159
• CPC Class Codes: G06F 18/2323 based on graph theory, e.g. minimum spanning trees [MST] or graph cuts
• CPC Class Codes: G06F 18/2431 Multiple classes
• CPC Class Codes: G06T 2207/10024 Color image
• CPC Class Codes: G06T 2207/20072 Graph-based image processing
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/162 involving graph-based methods
• CPC Class Codes: G06V 10/7635 based on graphs, e.g. graph cuts or spectral clustering

Abstract

A method for improving classification results of a classifier including receiving classification results for a plurality of elements that have been classified by a classifier as one of a plurality of classes, constructing a graph having a plurality of nodes, each node corresponding to one of the elements, and a plurality of labels, each label corresponding to one of the classes, adding edges between nodes corresponding to related elements, adding edges between each node and each label, and using a graph cut algorithm to cut edges to a node and partition the graph into classes, the graph cut algorithm using as input the classification results for the element corresponding to that node and related elements.

Claims

1. A method for improving classification results of a classifier including:

receiving, using a processor, classification results for a plurality of elements that have been classified by a classifier as one of a plurality of classes,

constructing, using a processor, a graph having a plurality of nodes, each node corresponding to one of the elements, and a plurality of labels, each label corresponding to one of the classes,

adding, using a processor, edges between nodes corresponding to related elements,

adding, using a processor, edges between each node and each label, and

using a graph cut algorithm to cut edges to a node and partition the graph into classes, the graph cut algorithm using as input the classification results for the element corresponding to that node and related elements.

2. A method according to claim 1, wherein the classifier is a multi-class classifier that has been applied to classify the elements as one of three or more classes.

3. A method according to claim 1, wherein the graph cut algorithm is a multi-class graph cut algorithm to partition the graph into three or more classes.

4. A method according to claim 1, wherein the classifier is a boosted classifier.

5. A method according to claim 1, wherein the classification results include a confidence value for each element, indicating confidence in the classification of that element, and the graph cut algorithm further uses as input the confidence value of the classification for the element corresponding to that node.

6. A method according to claim 5, wherein the graph cut algorithm further uses as input the confidence value of the classification for related elements.

7. A method according to claim 1, wherein the graph cut algorithm further uses as input a likelihood that related elements have the same classification.

8. A method according to claim 3, wherein the graph cut algorithm includes an alpha expansion procedure.

9. A method according to claim 1, wherein each element is a pixel in an image.

10. A method according to claim 9, wherein related elements are adjacent pixels in the image.

11. A method according to claim 10, wherein related elements are eight adjacent pixels.

12. A method according to claim 10, wherein related elements are four adjacent pixels.

13. A method according to claim 1, wherein each element is a contiguous region in an image.

14. A method according to claim 13, wherein related elements are adjacent contiguous regions in the image.

15. A method according to claim 13, wherein the classification results include a confidence value for each element, indicating confidence in the classification of that element, and the graph cut algorithm further uses as input the confidence value of the classification for the element corresponding to that node, and wherein each contiguous region includes pixels with the same classification and a confidence value above a threshold.

16. A method according to claim 9, wherein the graph cut algorithm further uses as input an expected size of an object in the image.

17. A method according to claim 9, wherein the graph cut algorithm further uses as input a predetermined rule regarding adjacent pixels.

18. (canceled)

19. A non-transitory computer readable media including software for use with a computer including a processor and memory for storing the software, the software including a series of instructions executable by the processor to carry out a method for improving classification results of a classifier, the method including:

receiving, using a processor, classification results for a plurality of elements that have been classified by a classifier as one of a plurality of classes,

constructing, using a processor, a graph having a plurality of nodes, each node corresponding to one of the elements, and a plurality of labels, each label corresponding to one of the classes,

adding, using a processor, edges between nodes corresponding to related elements,

adding, using a processor, edges between each node and each label, and using a graph cut algorithm to cut edges to a node and partition the graph into classes, the graph cut algorithm using as input the classification results for the element corresponding to that node and related elements.

20. Apparatus including

a processor;

a memory; and

software resident in memory accessible to the processor, the software including a series of instructions executable by the processor to carry out a method for improving classification results of a classifier, the method including;

receiving, using a processor, classification results for a plurality of elements that have been classified by a classifier as one of a plurality of classes,

constructing, using a processor, a graph having a plurality of nodes, each node corresponding to one of the elements, and a plurality of labels, each label corresponding to one of the classes,

adding, using a processor, edges between nodes corresponding to related elements,

adding, using a processor, edges between each node and each label, and

using a graph cut algorithm to cut edges to a node and partition the graph into classes, the graph cut algorithm using as input the classification results for the element corresponding to that node and related elements.