Multilevel thresholding for mutual information based registration and image registration using a GPU

1. A method of registering images, the method comprising:

for each image, determining an optimum intensity threshold set from a plurality of intensity threshold sets that maximizes a variance between classes of each set using a computer;

segmenting each image using the corresponding determined optimum intensity threshold set using the computer;

generating a joint histogram of at least two of the segmented images using the computer;

generating mutual information from the joint histogram using the computer; and

registering the at least two images based on the mutual information using the computer,

wherein the joint histogram is generated by a geometry shader of a graphical processing unit GPU of the computer in a texture object attached to a frame buffer of the GPU.

2. The method of claim 1 , wherein determining the optimum intensity threshold set for each image comprises:

generating a variance for each intensity threshold set; and

selecting the intensity threshold set that has the highest corresponding variance as the optimum intensity threshold set.

3. The method of claim 1 , wherein the thresholds of each intensity threshold set are pre-sorted and values of the thresholds of each intensity threshold set differ from one another.

4. The method of claim 2 , wherein generating a variance for each intensity threshold set comprises:

for each class,

determining a class occurrence probability from a probability distribution of intensity levels of the corresponding class;

determining a class mean level from the class occurrence probability;

determining a global mean from an entire range of intensity levels of the image and the probability distribution;

determining a value based on the class occurrence probability, the class mean level, and the global mean; and

summing the values to generate the variance.

5. The method of claim 4 , wherein the class occurrence probability is read out of a look up table.

6. The method of claim 4 , wherein determining the value comprises:

subtracting the global mean from the class mean level;

squaring a result of the subtraction; and

multiplying a result of the squaring by the class occurrence probability to generate the value.

7. The method of claim 4 , wherein determining the probability distribution of the intensity levels comprises:

determining a frequency of occurrence of each level in the image;

determining a probability for each frequency by dividing the frequency by the number of pixels of the image; and

summing all of the probabilities to generate the probability distribution.

8. The method of claim 4 , wherein determining the class mean level from the class occurrence probability comprises:

determining a value for each intensity level by multiplying the intensity level by the corresponding probability divided by the probability distribution; and

summing each of the values to generate the class mean level.

9. The method of claim 4 , wherein determining the global mean comprises:

determining a value for each intensity level in the entire range by multiplying each intensity level by a probability of the intensity level occurring in the image; and

summing each of the values to generate the global mean.

10. The method of claim 2 , wherein generating a variance for each intensity threshold set comprises:

for each class,

determining a class occurrence probability from a probability distribution of intensity levels of the corresponding class;

determining a class mean level from the class occurrence probability;

determining a value based on the class occurrence probability and the class mean level; and

summing the values to generate the variance.

11. The method of claim 10 , wherein determining the value comprises:

squaring the class mean level; and

multiplying a result of the squaring by the class occurrence probability to generate the value.

12. The method of claim 1 , wherein the generating of the mutual information is performed by a central processing unit CPU of the computer system separate from the GPU.

13. The method of claim 1 , wherein the registering is performed by a central processing unit CPU of the computer system separate from the GPU.

14. The method of claim 1 , wherein the generating of the mutual information and the registering are both performed by a central processing unit CPU of the computer system separate from the GPU.

15. The method of claim 1 , wherein the optimum intensity threshold set corresponds to a plurality of classes, where each class indicates an intensity range and a single intensity value, wherein the segmenting of each image comprises:

determining which of the classes each intensity in the segmented image belongs based on the intensity ranges; and

replacing each intensity in the segmented image with the single intensity value of its determined class.

16. The method of claim 15 , wherein at least one of the intensity ranges includes at least two different intensity values.

17. The method of claim 16 , wherein the intensity ranges are unique from one another.

18. The method of claim 17 , wherein the classes number at least three.

19. A computer program product to register images, said computer program product comprising:

a computer readable storage medium; and

program instructions to:

for each image, determine an optimum intensity threshold set from a plurality of intensity threshold sets that maximizes a variance between classes of each set;

segment each image using the corresponding determined optimum intensity threshold set;

generate a joint histogram of at least two of the segmented images;

generate mutual information from the joint histogram; and

register the at least two images based on the mutual information,

wherein the joint histogram is generated by a geometry shader of a graphical processing unit GPU in a texture object attached to a frame buffer of the GPU, and

wherein said program instructions are stored on said computer readable storage medium.

20. A method of registering images, the method comprising:

generating, by a geometry shader of a graphical processing unit GPU, a joint histogram of at least two segmented images in a texture object attached to a frame buffer of the GPU;

generating mutual information from the joint histogram; and

registering the at least two images based on the mutual information.

21. The method of claim 20 , wherein prior to the generating of the join histogram the method comprises:

for each image, determining an optimum intensity threshold set from a plurality of intensity threshold sets that maximizes a variance between classes of each set; and

segmenting each image using the corresponding determined optimum intensity threshold to generate the segmented images.

22. The method of claim 21 , wherein determining the optimum intensity threshold set for each image comprises:

generating a variance for each intensity threshold set; and

selecting the intensity threshold set that has the highest corresponding variance as the optimum intensity threshold set.

23. The method of claim 21 , wherein the thresholds of each intensity threshold set are pre-sorted and values of the thresholds of each intensity threshold set differ from one another.