Document clusterization using neural networks

1. A method, comprising:

detecting, by a processing device, a set of keypoints in an input image;

generating a set of keypoint vectors, wherein each keypoint vector of the set of keypoint vectors is associated with a corresponding keypoint of the set of keypoints;

extracting a feature map from the input image;

producing a combination of the set of keypoint vectors with the feature map;

transforming the combination into a set of keypoint mapping vectors according to a predefined mapping scheme;

estimating, based on the set of keypoint mapping vectors, a plurality of importance factors associated with the set of keypoints; and

classifying the input image based on the set of keypoints and the plurality of importance factors.

2. The method of claim 1 , wherein each keypoint vector of the set of keypoint vectors reflects a global context of the input image.

3. The method of claim 1 , wherein extracting the feature map from the input image further comprises:

applying a set of convolution filters to the input image.

4. The method of claim 1 , wherein producing the combination of the set of keypoint vectors with the feature map further comprises:

processing the combination by a neural network comprising one or more convolution layers.

5. The method of claim 1 , wherein producing the combination of the set of keypoint vectors with the feature map comprises at least one of: concatenating a first keypoint vector of the set of keypoint vectors with the feature map or summing a second keypoint vector of the set of keypoint vectors with the feature map.

6. The method of claim 1 , wherein estimating the plurality of importance factors associated with the set of keypoints further comprises:

selecting a subset of keypoints having their respective importance factors exceeding a chosen threshold importance value.

7. The method of claim 1 , wherein classifying the input image further comprises:

associating the input image with a cluster of images based on the set of keypoints and the plurality of importance factors.

8. A system comprising:

a memory; and

a processing device operatively coupled to the memory, the processing device to:

detect a set of keypoints in an input image;

generate a set of keypoint vectors, wherein each keypoint vector of the set of keypoint vectors is associated with a corresponding keypoint of the set of keypoints;

extract a feature map from the input image;

produce a combination of the set of keypoint vectors with the feature map;

transform the combination into a set of keypoint mapping vectors according to a predefined mapping scheme;

estimate, based on the set of keypoint mapping vectors, a plurality of importance factors associated with the set of keypoints; and

classify the input image based on the set of keypoints and the plurality of importance factors.

9. The system of claim 8 , wherein each keypoint vector of the set of keypoint vectors reflects a global context of the input image.

10. The system of claim 8 , wherein extracting the feature map from the input image further comprises:

applying a set of convolution filters to the input image.

11. The system of claim 8 , wherein producing the combination of the set of keypoint vectors with the feature map further comprises:

processing the combination by a neural network comprising one or more convolution layers.

12. The system of claim 8 , wherein producing the combination of the set of keypoint vectors with the feature map comprises at least one of: concatenating a first keypoint vector of the set of keypoint vectors with the feature map or summing a second keypoint vector of the set of keypoint vectors with the feature map.

13. The system of claim 8 , wherein estimating the plurality of importance factors associated with the set of keypoints further comprises:

selecting a subset of keypoints having their respective importance factors exceeding a chosen threshold importance value.

14. The system of claim 8 , wherein classifying the input image further comprises:

associating the input image with a cluster of images based on the set of keypoints and the plurality of importance factors.

15. A non-transitory computer-readable storage medium including executable instructions that, when executed by a computing system, cause the computing system to:

detect a set of visual words in an input image;

generate a set of visual word vectors, wherein each visual word vector of the set of visual word vectors is associated with a corresponding visual word of the set of visual words;

extract a feature map from the input image;

produce a combination of the set of visual word vectors with the feature map;

transform the combination into a set of visual word mapping vectors according to a predefined mapping scheme;

estimate, based on the set of visual word mapping vectors, a plurality of importance factors associated with the set of visual words; and

classify the input image based on the set of visual words and the plurality of importance factors.

16. The non-transitory computer-readable storage medium of claim 15 , wherein each visual word vector of the set of visual word vectors reflects a global context of the input image.

17. The non-transitory computer-readable storage medium of claim 15 , wherein producing the combination of the set of visual word vectors with the feature map further comprises:

processing the combination by a neural network comprising one or more convolution layers.

18. The non-transitory computer-readable storage medium of claim 15 , wherein producing the combination of the set of visual word vectors with the feature map comprises at least one of: concatenating a first visual word vector of the set of visual word vectors with the feature map or summing a second visual word vector of the set of visual word vectors with the feature map.

19. The non-transitory computer-readable storage medium of claim 15 , wherein estimating the plurality of importance factors associated with the set of visual words further comprises:

selecting a subset of visual words having their respective importance factors exceeding a chosen threshold importance value.

20. The non-transitory computer-readable storage medium of claim 15 , wherein classifying the input image further comprises:

associating the input image with a cluster of images based on the set of visual words and the plurality of importance factors.