Method and apparatus for generating learning data required to learn animation characters based on deep learning

1. A learning data generation method to learn animation characters on the basis of deep learning, the learning data generation method comprising:

collecting various images from an external source using wired/wireless communication;

training a character detection module using the collected images;

acquiring character images from the collected images using the trained character detection module;

clustering the acquired character images;

selecting learning data from among the clustered images; and

inputting the selected learning data to an artificial neural network for character recognition,

wherein the training of the character detection module comprises:

labeling the collected images to generate labels corresponding to the respective images; and

inputting the generated labels and the collected images to a preset character detection model to train the character detection model.

2. The learning data generation method of claim 1 , wherein the collecting of the various images comprises:

collecting a video from the external source using the wired/wireless communication; and

extracting frames from the collected video at preset time intervals.

3. The learning data generation method of claim 1 , wherein each of the labels is a label conforming to DarkNet format and includes a class number, a start point coordinate ratio of a bounding box, and an aspect ratio of the bounding box.

4. The learning data generation method of claim 1 , wherein the acquiring of the character images comprises:

detecting a bounding-box-shaped character region from each of the collected images using the trained character detection model; and

preprocessing the detected character region to acquire the corresponding character image.

5. The learning data generation method of claim 4 , wherein the preprocessing comprises cutting the character region out of the corresponding collected image and normalizing the size of the cut character region.

6. The learning data generation method of claim 1 , wherein the clustering comprises:

extracting a feature vector from the character image; and

clustering the character image using the extracted feature vector.

7. The learning data generation method of claim 6 , wherein the clustering of the character image using the extracted feature vector comprises:

clustering the character image;

adjusting a cluster determination strength value according to the number of clusters generated by the clustering; and

re-clustering the character image.

8. The learning data generation method of claim 7 , wherein the adjusting of the cluster determination strength value comprises:

increasing the cluster determination strength value when the number of clusters exceeds a maximal limit value within a preset limit range; and

decreasing the cluster determination strength value when the number of clusters is less than an initial limit value within the limit range.

9. A learning data generation apparatus needed to learn animation characters on the basis of deep learning, the learning data generation apparatus comprising:

at least one processor; and

a memory configured to store instructions for instructing the at least one processor to perform at least one operation,

wherein the at least one operation comprises:

collecting various images from an external source using wired/wireless communication;

training a character detection module using the collected images;

acquiring character images from the collected images using the trained character detection module;

clustering the acquired character images;

selecting learning data from among the clustered images; and

inputting the selected learning data to an artificial neural network for character recognition,

wherein the training of the character detection module comprises:

labeling the collected images to generate labels corresponding to the respective images; and

inputting the generated labels and the collected images to a preset character detection model to train the character detection model.

10. The learning data generation apparatus of claim 9 , wherein the collecting of the various images comprises:

collecting a video from the external source using the wired/wireless communication; and

extracting frames from the collected video at preset time intervals.

11. The learning data generation apparatus of claim 9 , wherein the character detection model is one of You-Only-Look-Once (YOLO), Single Shot Detector (SSD), and Faster R-convolutional neural network (CNN).

12. The learning data generation apparatus of claim 9 , wherein each of the labels is a label conforming to DarkNet format and includes a class number, a start point coordinate ratio of a bounding box, and an aspect ratio of the bounding box.

13. The learning data generation apparatus of claim 9 , wherein the acquiring of the character images comprises:

detecting a bounding-box-shaped character region from the collected image using the trained character detection model; and

preprocessing the detected character region to acquire the character image.

14. The learning data generation apparatus of claim 13 , wherein the preprocessing comprises cutting the character region out of the corresponding collected image and normalizing the size of the cut character region.

15. The learning data generation apparatus of claim 9 , wherein the clustering comprises:

extracting a feature vector from the character image; and

clustering the character image using the extracted feature vector.

16. The learning data generation apparatus of claim 9 , wherein the selecting of the learning data comprises:

displaying, to a user, representative images for a plurality of clusters to which the clustered image belongs; and

selecting, as the learning data, a character image belonging to a cluster corresponding to an image selected by the user from among the displayed representative images.