GPU multi-label image segmentation

• US 20050226506A1
• Filed: 04/04/2005
• Published: 10/13/2005
• Est. Priority Date: 04/09/2004
• Status: Active Grant

• Current Assignee: Siemens Medical Solutions USA Incorporated (Siemens AG)
• Original Assignee: Siemens Medical Solutions USA Incorporated (Siemens AG)
• Inventors: Aharon, Shmuel, Grady, Leo
• Granted Patent: US 7,697,756 B2
• Time in Patent Office: Days
• US Class Current: 382/180
• CPC Class Codes: G06T 2200/28 involving image processing hardware
• CPC Class Codes: G06T 2207/20072 Graph-based image processing
• CPC Class Codes: G06T 2207/20101 Interactive definition of point of interest, landmark or seed
• CPC Class Codes: G06T 2207/30004 Biomedical image processing
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling

Abstract

A method for image segmentation includes specifying seed points in an image of interest, the seed points corresponding to a node in a seed texture, each seed point having a different color. The method includes determining a matrix for each node, including neighboring edge weights of each node, and determining a probability that a node can be characterized as each seed point. The method includes assigning the node the color of a most probable seed point, and outputting a segmentation of the image of interest according to node assignments, wherein the segmentation differentiates portions of the image of interest.

Claims

1. A method for image segmentation comprising:

specifying a plurality of seed points in an image of interest, the seed points corresponding to a node in a seed texture, each seed point having a different color;

determining a matrix for each node, including neighboring edge weights of each node;

determining a probability that a node can be characterized as each seed point;

assigning the node the color of a most probable seed point; and

outputting a segmentation of the image of interest according to node assignments, wherein the segmentation differentiates portions of the image of interest.

2. The method of claim 1, further comprising determining edge weights between each node in the seed texture.

3. The method of claim 1, wherein determining the matrix comprises determining a Laplacian matrix having five diagonal bands, wherein four secondary bands hold the edge weights and a main band is a sum of the four secondary bands.

4. The method of claim 3, further comprising storing the Laplcian matrix of edge weights as a texture of RGBA channels.

5. The method of claim 3, further comprising:

determining the sum for each node; and

determining a vector of the sums for each color, the colors being RGBA channels.

6. The method of claim 5, wherein determining the sum for each node further comprising determining a dot product of the neighbors for each node.

7. The method of claim 1, further comprising determining the probabilities by conjugate gradient vector.

8. A program storage device readable by machine, tangibly embodying a program of instructions executable by the machine to perform a method for image segmentation, the method comprising:

specifying a plurality of seed points in an image of interest, the seed points corresponding to a node in a seed texture, each seed point having a different color;

determining a matrix for each node, including neighboring edge weights of each node;

determining a probability that a node can be characterized as each seed point;

assigning the node the color of a most probable seed point; and

outputting a segmentation of the image of interest according to node assignments, wherein the segmentation differentiates portions of the image of interest.

9. The method of claim 8, further comprising determining edge weights between each node in the seed texture.

10. The method of claim 8, wherein determining the matrix comprises determining a Laplacian matrix having five diagonal bands, wherein four secondary bands hold the edge weights and a main band is a sum of the four secondary bands.

11. The method of claim 10, further comprising storing the Laplcian matrix of edge weights as a texture of RGBA channels.

12. The method of claim 10, further comprising:

determining the sum for each node; and

determining a vector of the sums for each color, the colors being RGBA channels.

13. The method of claim 12, wherein determining the sum for each node further comprising determining a dot product of the neighbors for each node.

14. The method of claim 8, further comprising determining the probabilities by conjugate gradient vector.

15. A graphics processing unit coupled to a bus of a computer system for processing a Laplacian matrix of edge weights between nodes, each node corresponding to a pixel in an image of interest, the graphics processing unit for determining an output texture comprising vector data for each RGBA channel and a label for each node, wherein the label corresponds to an RGBA channel, the graphics processing unit for outputting a segmentation of the image of interest according to the label for each node.

16. The graphics processing unit of claim 15, wherein the Laplacian matrix is a texture representation of the edge weights between nodes stored in a memory device, the memory device being on the graphics processing unit or connected to the bus.