GPU accelerated multi-label image segmentation (MLS)

1. An image segmentation device comprising:

a memory device storing an image of interest and a plurality of instructions for segmenting the image of interest; and

a graphics processing unit for receiving the image of interest and executing the plurality of instructions to perform a method comprising,

specifying a plurality of seed points in the image of interest;

determining a graph of nodes representing the image, wherein each node corresponds to a pixel of the image and neighboring edge weights between neighboring nodes represent differences in image intensities between neighboring pixels, wherein determining the graph 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 and storing the Laplacian matrix of edge weights as a texture representation having a plurality of channels;

determining a vector texture of vector data representing different potential labels of the nodes, the vector data for each label is determined by matrix-vector multiplication of the secondary diagonals, the main band, and a sample vector for each node from the first texture;

determining a probability that a node of the graph belongs to each potential label, wherein the probabilities are determined for each node in parallel as a conjugate gradient vector of the vector texture;

assigning each node a most probable label based on the probabilities; and

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

2. The image segmentation device of claim 1 , further comprising determining edge weights between neighboring nodes in the graph.

3. The image segmentation device of claim 1 , further comprising:

determining the sum for each node; and

determining a vector of the sums for each channel, the channel being colors associated with the potential labels.

4. The image segmentation device of claim 3 , wherein determining the sum for each node further comprising determining a dot product of the neighbors for each node.

5. The image segmentation device of claim 1 , further comprising determining the probabilities by conjugate gradient vector.

6. A computer readable medium embodying instructions executable by a processor to perform a method for image segmentation, the method comprising:

specifying a plurality of seed points in an image of interest;

determining a graph of nodes representing the image, wherein each node corresponds to a pixel of the image and neighboring edge weights between neighboring nodes represent differences in image intensities between neighboring pixels, wherein determining the graph 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 and storing the Laplacian matrix of edge weights as a texture representation having a plurality of channels;

determining a vector texture of vector data representing different potential labels of the nodes, the vector data for each label is determined by matrix-vector multiplication of the secondary diagonals, the main band, and a sample vector for each node from the first texture;

determining a probability that a node of the graph belongs to each potential label, wherein the probabilities are determined for each node in parallel as a conjugate gradient vector of the vector texture;

assigning each node a most probable label based on the probabilities; and

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

7. The computer readable medium of claim 6 , further comprising determining edge weights between neighboring nodes in the graph.

8. The computer readable medium of claim 6 , further comprising:

determining the sum for each node; and

determining a vector of the sums for each channel, the channel being colors associated with the potential labels.

9. The computer readable medium of claim 8 , wherein determining the sum for each node further comprising determining a dot product of the neighbors for each node.

10. The computer readable medium of claim 6 , further comprising determining the probabilities by conjugate gradient vector.