Devices, systems, and methods for clustering reference images for non-destructive testing
Devices, systems, and methods obtain training images; generate image-alignment data based on the training images; cluster the training images based at least in part on the image-alignment data, thereby generating clusters of training images; and select one or more representative images from the training images based on the clusters of training images.
1. A device comprising:
one or more computer-readable storage media; and
one or more processors that are configured to cause the device to perform operations including:
obtaining training images;
generating image-alignment data based on the training images, wherein generating the image-alignment data includes generating respective transformations between pairs of images of the training images;
clustering the training images based at least in part on the image-alignment data, thereby generating clusters of training images; and
selecting one or more representative images from the training images based on the clusters of training images.
2. The device of claim 1 , wherein generating the image-alignment data includes:
selecting a target-alignment image of the training images, and
aligning the other images of the training images to the target-alignment image, thereby generating aligned training images.
3. The device of claim 1 , wherein the transformations are affine transformations that transforms point (x, y) to point (x′, y′) as described by
[
x
′
y
′
1
]
=
T
[
x
y
1
]
=
[
A
❘
δ
y
δ
x
0
0
❘
1
]
[
x
y
1
]
,
where A is a 2×2 matrix; where and δ x and δ y are translation amounts; and where T is a 3×3 affine transformation matrix that can be described by matrix A, a two-dimensional translation column vector δ, and a constant row of zeros with a one at the bottom-right corner.
4. The device of claim 1 , wherein generating the image- alignment data includes:
generating
(
N
2
)
transformations, where N is the number of training images.
5. The device of claim 1 , wherein the operations further include:
generating reduced-dimensionality representations of the image-alignment data, and
wherein clustering the training images is based on the reduced-dimensionality representations of the image-alignment data.
6. The device of claim 1 , wherein selecting the one or more representative images from the training images based on the clusters of training images includes:
selecting a respective centroid image or a respective median image from one or more clusters of the clusters of training images as a representative image of a cluster of the one or more clusters.
7. The device of claim 1 , wherein the operations further include:
calculating dissimilarity measurements between the training images,
wherein clustering the training images is further based on the dissimilarity measurements.
8. The device of claim 7 , wherein each of the dissimilarity measurements measures a dissimilarity between a respective pair of images of the training images.
9. A method comprising:
obtaining training images;
generating image-alignment data based on the training images, wherein generating the image-alignment data includes generating respective transformations between pairs of images of the training images;
clustering the training images based at least in part on the image-alignment data, thereby generating clusters of training images; and
selecting representative images from the training images based on the clusters of training images.
10. The method of claim 9 , wherein generating the image-alignment data includes:
selecting a target-alignment image of the training images, and
aligning the other images of the training images to the target-alignment image, thereby generating aligned training images.
11. The method of claim 9 , further comprising:
calculating dissimilarity measurements between the training images,
wherein clustering the training images is further based on the dissimilarity measurements.
12. The method of claim 11 , further comprising:
reducing a dimensionality of the dissimilarity measurements, thereby generating reduced-dimensionality dissimilarity measurements,
wherein clustering the training images is further based on the reduced- dimensionality dissimilarity measurements.
13. The method of claim 12 , wherein selecting the representative images from the training images based on the clusters of training images includes:
selecting at least some of the training images that are in the clusters of training images that include more than a threshold number of training images as the representative images.
14. One or more computer-readable media storing instructions that, when executed by one or more computing devices, cause the one or more computing devices to execute operations comprising:
obtaining training images;
generating image-alignment data based on the training images, wherein the image-alignment data include pair-wise transformations between pairs of the training images;
clustering the training images based at least in part on the image-alignment data, thereby generating clusters of training images; and
selecting representative images from the training images based on the clusters of training images.
15. The one or more computer-readable media of claim 14 , wherein the image-alignment data include respective pair-wise transformations between a target-alignment image of the training images and other images of the training images.
16. The one or more computer-readable media of claim 14 , wherein selecting the representative images from the training images based on the clusters of training images includes:
selecting a respective centroid image or a respective median image from a cluster of the clusters of training images as a representative image of the cluster.
17. A device comprising:
one or more computer-readable storage media; and
one or more processors that are configured to cause the device to perform operations including:
obtaining training images;
generating image-alignment data based on the training images;
generating reduced-dimensionality representations of the image-alignment data;
clustering the training images based at least in part on the reduced- dimensionality representations of the image-alignment data, thereby generating clusters of training images; and
selecting one or more representative images from the training images based on the clusters of training images.
18. A method comprising:
obtaining training images;
generating image-alignment data based on the training images;
calculating dissimilarity measurements between the training images;
reducing a dimensionality of the dissimilarity measurements, thereby generating reduced-dimensionality dissimilarity measurements;
clustering the training images based at least in part on the image-alignment data and at least in part on the reduced-dimensionality dissimilarity measurements, thereby generating clusters of training images; and
selecting representative images from the training images based on the clusters of training images.
19. One or more computer-readable media storing instructions that, when executed by one or more computing devices, cause the one or more computing devices to execute operations comprising:
obtaining training images;
generating image-alignment data based on the training images;
calculating dissimilarity measurements between the training images based on the image-alignment data;
reducing a dimensionality of the dissimilarity measurements, thereby generating reduced-dimensionality dissimilarity measurements;
clustering the training images based at least in part on the reduced-dimensionality dissimilarity measurements, thereby generating clusters of training images; and
selecting representative images from the training images based on the clusters of training images.
20. The device of claim 2 , wherein the respective transformations between the pairs of images of the training images include
respective transformations between the target-alignment image and each of the other images of the training images.
21. The device of claim 7 , wherein the dissimilarity measurements are calculated based on the image-alignment data.