System and method for multi-label image segmentation
View Patent ↗A system and method for multi-label image segmentation is provided. The method comprises the steps of: receiving image data including a set of labeled image elements; mapping a change in intensities of the image data to edge weights; determining potentials for each image element in the image data; and assigning a label, based upon the determined potentials, to each image element in the image data.
1. A method for multi-label image segmentation, comprising:
receiving image data including a set of labeled image elements;
mapping a change in intensities of the image data to edge weights;
determining potentials for each image element in the image data based on the edge weights; and
assigning a label from a set of labels, based upon the determined potentials, to each image element in the image data;
wherein the label assigned to each image element corresponds to max s (x x i )
where max s is to maximize over variable s which takes value from the set of labels, x x i is the potential at an image element i corresponding to a label s.
2. The method of claim 1 , wherein the image data is marked by one of a user and a computer.
3. The method of claim 1 , wherein the image elements are one of a pixel and a voxel.
4. The method of claim 1 , wherein the image data includes unlabeled image elements.
5. The method of claim 1 , wherein the change in intensities of the image data to edge weights is mapped by a Gaussian weighting function.
6. The method of claim 5 , wherein the Gaussian weighting function is given by
w ij =exp(−β( g i −g j ) 2 ),
where g i indicates image intensity at image element i, g j indicates image intensity at image element j, and β is a free parameter.
7. The method of claim 1 , wherein the change in intensities of the image data to edge weights is mapped to represent the image data with random walker biases.
8. The method of claim 1 , wherein the potentials for each image element in the image data are determined by
L U X=−BM,
where L U is a reduced Laplacian matrix, X is a set of probabilities for each image element in the image data, B is a joining block between labeled and unlabeled image elements in a Laplacian matrix and M is a set of indicator values for indicating values of the labeled image elements.
9. The method of claim 1 , wherein the determined potentials represent the probability that a random walker starting at an image element in the image data first reaches a seed point in the image data when the seed point is set to unity.
10. The method of claim 1 , wherein the labels assigned to each image element are represented by a color.
11. The method of claim 1 , further comprising:
outputting the assigned label.
12. The method of claim 1 , further comprising:
acquiring the image data.
13. The method of claim 12 , wherein the image data is acquired by one of a computed tomographic (CT), helical CT, x-ray, positron emission tomographic (PET), fluoroscopic, ultrasound, and magnetic resonance (MR) imaging technique.
14. The method of claim 12 , wherein the image data is acquired by one of a digital camera and a flatbed scanner.
15. A system for multi-label image segmentation, comprising:
a memory device for storing a program;
a processor in communication with the memory device, the processor operative with the program to:
receive image data including a set of labeled image elements;
map a change in intensities of the image data to edge weights;
determine potentials for each image element in the image data base don the edge weights; and
assign a label from a set of labels, based upon the determined potentials, to each image element in the image data,
wherein the label assigned to each image element corresponds to max s (x i s )
where max s is to maximize over variable s which takes value from the set of labels, x i s is the potential at an image element i corresponding to a lable s.
16. The system of claim 15 , wherein the image data is marked by one of a user and a computer.
17. The system of claim 15 , wherein the image elements are one of a pixel and a voxel.
18. The system of claim 15 , wherein the image data includes unlabeled image elements.
19. The system of claim 15 , wherein the change in intensities of the image data to edge weights are mapped by a Gaussian weighting function to represent the image data with random walker biases.
20. The system of claim 19 , wherein the Gaussian weighting function is given by
w ij =exp( g i −g j ) 2 ),
where g i indicates image intensity at image element i, g j indicates image intensity at image element j, and β is a free parameter.
21. The system of claim 19 , wherein the random walker is biased to avoid crossing object boundaries.
22. The system of claim 15 , wherein the potentials for each image element in the image data are determined by
L U X=−BM,
where L U is a reduced Laplacian matrix, X is a set of probabilities for each image element in the image data, B is a joining block between labeled and unlabeled image elements in a Laplacian matrix and M is a set of indicator values for indicating values of the labeled image elements.
23. The system of claim 15 , wherein the determined potentials represent the probability that a random walker starting at an image element in the image data first reaches a seed point in the image data when the seed point is set to unity.
24. The system of claim 15 , wherein the labels assigned to each image element are represented by a color.
25. The system of claim 15 , wherein the processor is further operative with the program code to:
output the assigned label.
26. The system of claim 15 , wherein the processor is further operative with the program code to:
acquire the image data.
27. The system of claim 26 , wherein the image data is acquired by one of a computed tomographic (CT), helical CT, x-ray, positron emission tomographic (PET), fluoroscopic, ultrasound, and magnetic resonance (MR) imaging technique.
28. The system of claim 26 , wherein the image data is acquired by one of a digital camera and a flatbed scanner.
29. A computer readable storage medium storing computer program code that, when executed by a computer, causes the computer to perform multi-label image segmentation, the computer program logic comprising:
program code for receiving image data including a set of labeled image elements;
program code for mapping a change in intensities of the image data to edge weights;
program code for determining potentials for each image element in the image data based on the edge weights; and
program code for assigning a label from a set of labels, based upon the determined potentials, to each image element in the image data,
wherein the label assigned to each image element corresponds to max s(x i s )
where max s is to maximize over variable s which takes value from the set of lables, x j s is the potential at an image element i corresponding to a label s.
30. A system for multi-label image segmentation, comprising:
means for receiving image data including a set of labeled image elements;
means for mapping a change in intensities of the image data to edge weights;
means for determining potentials for each image element in the image data based on the edge weights; and
means for assigning a label form a set of labels, based upon the determined potentials, to each image element in the image data,
wherein the label assigned to each image element corresponds to max s (x i s )
where max s is to maximize over variable s which takes value from the set of lables, x j s is the potential at an image element i corresponding to a label s.
31. A method for semi-automatic multi-label image segmentation, comprising:
acquiring image data;
marking the image data to include a set of labeled image elements;
mapping a change in intensities of the image data to edge weights to represent the image data with random walker biases;
determining potentials for each image element in the image databased on the edge weights, wherein the determined potentials represent the probability that a random walker starting at an image element in the image data first reaches a seed point in the image data set to unity;
assigning a label from a set of labels, based upon the determined potentials, to each image element in the image data; and
outputting a segmentation of the image data including the assigned label s,
wherein the label assigned to each image element corresponds to max s (x i s)
where max s is to maximize over variable s which takes value from the set of lables, x j s is the potential at an image element i corresponding to a label s.