Recurrent pattern image classification and registration

1. A method comprising:

receiving a source image and a target image, each depicting a dense recurring pattern comprising an array of objects arranged in close proximity to one another;

applying a trained machine learning classifier to obtain a classification of each pixel in said source image and said target image into one of at least two classes;

determining a pixel-level transformation between said classified source and target images, based, at least in part, on a set of transformation parameters;

training a neural network to optimize said set of transformation parameters, based, at least in part, on minimizing a loss function which calculates a weighted sum of per-pixel matches between said classified source and target images; and

applying said optimized set of transformation parameters to said target image to align said target image with said source image.

2. The method of claim 1 , wherein said minimizing is based, at least in part, on a gradient descent algorithm.

3. The method of claim 1 , wherein at least some of said objects are in contact engagement with one another.

4. The method of claim 1 , wherein said objects are selected from the group consisting of trees, plants, shrubs, and crops, and wherein said array comprises a planting layout of said objects in one of an orchard, a grove, a field, and a wood.

5. The method of claim 1 , wherein said source and target images are selected from the group consisting of: images taken concurrently, images taken consecutively, and images depicting geographical areas which overlap at least partially.

6. The method of claim 1 , wherein said machine learning classifier is trained on a training set comprising (i) a plurality of images depicting a dense recurring pattern, and (ii) labels associated with said at least two classes, and wherein said images are labelled with said labels.

7. The method of claim 1 , wherein said set of classes comprises at least some of foliage, tree trunk, tree limb, grass, undergrowth, road, gravel path, alley, rocky formation, and stony formation.

8. A system comprising:

at least one hardware processor; and

a non-transitory computer-readable storage medium having stored thereon program instructions, the program instructions executable by the at least one hardware processor to:

receive a source image and a target image, each depicting a dense recurring pattern comprising an array of objects arranged in close proximity to one another,

apply a trained machine learning classifier to obtain a classification of each pixel in said source image and said target image into one of at least two classes,

determine a pixel-level transformation between said classified source and target images, based, at least in part, on a set of transformation parameters,

train a neural network to optimize said set of transformation parameters, based, at least in part, on minimizing a loss function which calculates a weighted sum of per-pixel matches between said classified source and target images, and

apply said optimized set of transformation parameters to said target image to align said target image with said source image.

9. The system of claim 8 , wherein said minimizing is based, at least in part, on a gradient descent algorithm.

10. The system of claim 8 , wherein at least some of said objects are in contact engagement with one another.

11. The system of claim 8 , wherein said objects are selected from the group consisting of trees, plants, shrubs, and crops, and wherein said array comprises a planting layout of said objects in one of an orchard, a grove, a field, and a wood.

12. The system of claim 8 , wherein said source and target images are selected from the group consisting of: images taken concurrently, images taken consecutively, and images depicting geographical areas which overlap at least partially.

13. The system of claim 8 , wherein said machine learning classifier is trained on a training set comprising (i) a plurality of images depicting a dense recurring pattern, and (ii) labels associated with said at least two classes, and wherein said images are labelled with said labels.

14. The system of claim 8 , wherein said set of classes comprises at least some of foliage, tree trunk, tree limb, grass, undergrowth, road, gravel path, alley, rocky formation, and stony formation.

15. A computer program product comprising a non-transitory computer-readable storage medium having program instructions embodied therewith, the program instructions executable by at least one hardware processor to:

receive a source image and a target image, each depicting a dense recurring pattern comprising an array of objects arranged in close proximity to one another;

apply a trained machine learning classifier to obtain a classification of each pixel in said source image and said target image into one of at least two classes;

determine a pixel-level transformation between said classified source and target images, based, at least in part, on a set of transformation parameters;

train a neural network to optimize said set of transformation parameters, based, at least in part, on minimizing a loss function which calculates a weighted sum of per-pixel matches between said classified source and target images; and

apply said optimized set of transformation parameters to said target image to align said target image with said source image.

16. The computer program product of claim 15 , wherein said minimizing is based, at least in part, on a gradient descent algorithm.

17. The computer program product of claim 15 , wherein at least some of said objects are in contact engagement with one another.

18. The computer program product of claim 15 , wherein said objects are selected from the group consisting of trees, plants, shrubs, and crops, and wherein said array comprises a planting layout of said objects in one of an orchard, a grove, a field, and a wood.

19. The computer program product of claim 15 , wherein said source and target images are selected from the group consisting of: images taken concurrently, images taken consecutively, and images depicting geographical areas which overlap at least partially.

20. The computer program product of claim 15 , wherein said machine learning classifier is trained on a training set comprising (i) a plurality of images depicting a dense recurring pattern, and (ii) labels associated with said at least two classes, and wherein said images are labelled with said labels.