Defective pixel identification using machine learning

1. One or more non-transitory computer readable media embodying a program of instructions executable by machine to perform operations for defective pixel identification, the operations comprising:

receiving a training image dataset;

training, using the training image dataset, convolutional neural networks to identify defective pixels;

identifying one or more defective pixels in a current image by applying the trained convolutional neural networks to the current image and generating likelihood values; and

correcting the one or more identified defective pixels, and

wherein the training image dataset comprises raw data images and ground truth images, wherein the raw data images include images corrected for offset and gain only, and the ground truth images include images corrected for offset, gain and defective pixels.

2. The one or more non-transitory computer readable media of claim 1 wherein the one or more defective pixels appear in the current image as one or more isolated spots.

3. The one or more non-transitory computer readable media of claim 1 wherein the one or more defective pixels appear in the current image as one or more in-line patterns.

4. The one or more non-transitory computer readable media of claim 1 wherein the training image dataset comprises two-dimensional flat panel radiographic images.

5. A system comprising:

a non-transitory memory device for storing computer readable program code; and

a processor in communication with the memory device, the processor being operative with the computer readable program code to perform operations including

receiving a training image dataset,

performing a machine learning technique to train a classifier using the training image dataset,

identifying one or more defective pixels in a current image by applying the trained classifier to the current image, and

correcting the one or more identified defective pixels, and

wherein the training image dataset comprises raw data images and ground truth images, wherein the raw data images include images corrected for offset and gain only, and the ground truth images include images corrected for offset, gain and defective pixels.

6. The system of claim 5 wherein the training image dataset comprises two-dimensional flat panel radiographic images.

7. The system of claim 5 wherein the processor is operative with the computer readable program code to perform the machine learning technique based on random forests.

8. The system of claim 7 wherein the processor is operative with the computer readable program code to perform the machine learning technique by

determining spatial features to identify defective pixels,

growing decision trees from the training dataset, and

pruning the decision trees for fitness.

9. The system of claim 8 wherein the spatial features comprise pixel values, derivatives in x-direction or y-direction, median positions in a square neighborhood, or a combination thereof.

10. The system of claim 8 wherein the processor is operative with the computer readable program code to prune the decision trees for fitness by

randomly selecting at least one of the spatial features at a decision junction of at least one of the decision trees, and

determining an optimal threshold value of the spatial feature to satisfy ground truth.

11. The system of claim 5 wherein the processor is operative with the computer readable program code to perform the machine learning technique based on convolutional neural networks.

12. The system of claim 11 wherein the convolutional neural networks comprise one or more input layers, hidden layers and output layers.

13. The system of claim 12 wherein the one or more input layers comprise neurons that represent a patch of image pixels centered at a specific location.

14. The system of claim 12 wherein the one or more output layers comprise a node that provides results of the one or more hidden layers.

15. The system of claim 14 wherein the results represent a likelihood that a center pixel of a square patch of the current image is defective.

16. The system of claim 5 wherein the processor is operative with the computer readable program code to identify the one or more defective pixels by determining that a likelihood value generated by the trained classifier exceeds a predetermined threshold value.

17. The system of claim 5 wherein the processor is operative with the computer readable program code to correct the one or more identified defective pixels by replacing the one or more identified defective pixels with one or more interpolated values based on neighboring non-defective pixels.

18. The system of claim 17 wherein the processor is operative with the computer readable program code to generate the one or more interpolated values by performing linear interpolation of image intensity values from the neighboring non-defective pixels.

19. A method, comprising:

receiving a training image dataset;

performing a machine learning technique to train a classifier using the training image dataset;

identifying one or more defective pixels in a current image by applying the trained classifier; and

correcting the one or more identified defective pixels, and

wherein the training image dataset comprises raw data images and ground truth images, wherein the raw data images include images corrected for offset and gain only, and the ground truth images include images corrected for offset, gain and defective pixels.

20. A system comprising:

a non-transitory memory device for storing computer readable program code; and

a processor in communication with the memory device, the processor being operative with the computer readable program code to perform operations including

receiving a training image dataset;

performing a machine learning technique to train a classifier using the training image dataset, wherein the processor is operative with the computer readable program code to perform the machine learning technique based on random forests by

determining spatial features to identify defective pixels,

growing decision trees from the training dataset, and

pruning the decision trees for fitness;

identifying one or more defective pixels in a current image by applying the trained classifier to the current image; and

correcting the one or more identified defective pixels.

21. The system of claim 20 wherein the spatial features comprise pixel values, derivatives in x-direction or y-direction, median positions in a square neighborhood, or a combination thereof.

22. The system of claim 20 wherein the processor is operative with the computer readable program code to prune the decision trees for fitness by

randomly selecting at least one of the spatial features at a decision junction of at least one of the decision trees, and

determining an optimal threshold value of the spatial feature to satisfy ground truth.

23. A system comprising:

a non-transitory memory device for storing computer readable program code; and

a processor in communication with the memory device, the processor being operative with the computer readable program code to perform operations including

receiving a training image dataset,

performing a machine learning technique to train a classifier using the training image dataset, wherein the processor is operative with the computer readable program code to perform the machine learning technique based on convolutional neural networks comprising one or more input layers, hidden layers and output layers, wherein the one or more output layers comprise a node that provides results of the one or more hidden layers and the results represent a likelihood that a center pixel of a square patch of the current image is defective,

identifying one or more defective pixels in a current image by applying the trained classifier to the current image, and

correcting the one or more identified defective pixels.