Image classification and labeling
A method of training an image classification model includes obtaining training images associated with labels, where two or more labels of the labels are associated with each of the training images and where each label of the two or more labels corresponds to an image classification class. The method further includes classifying training images into one or more classes using a deep convolutional neural network, and comparing the classification of the training images against labels associated with the training images. The method also includes updating parameters of the deep convolutional neural network based on the comparison of the classification of the training images against the labels associated with the training images.
1. A method of classifying images using one or more image classification models, comprising:
obtaining, by processing circuitry, training images associated with labels, wherein at least one training image of the training images is associated with two or more labels of the labels, each label corresponding to an image classification class, the labels having a hierarchical structure;
training, by the processing circuitry, at least two convolutional neural networks using the training images and the hierarchically-structured labels associated with the training images, a separate convolutional neural network being trained for each level of the hierarchical structure; and
classifying, by the processing circuitry, an input image into two or more classes based on the trained at least two convolutional neural networks, the classifying the input image including
multiplying a probability score output from a first trained convolutional neural network for a label by a probability score output from a second trained convolutional neural network for the label.
2. The method according to claim 1 , wherein a classification layer of each respective convolutional neural network is based on soft-sigmoid activation, the soft-sigmoid activation being a combination of a softmax function and a sigmoid function.
3. The method according to claim 1 , wherein the training images and the input images include graphically-designed images.
4. The method according to claim 1 , wherein the labels are non-mutually exclusive labels.
5. The method according to claim 1 , wherein the labels are codes used by a trademark registration organization.
6. The method according to claim 1 , wherein the labels are codes used to classify design patent images or industrial design images.
7. The method according to claim 1 , wherein the labels are available as metadata of the training images associated with the labels.
8. The method according to claim 1 , wherein the classifying further includes
labelling the input image with two or more labels corresponding to the two or more classes.
9. The method according to claim 1 , further comprising
pre-processing, by the processing circuitry, the training images, wherein the training the at least two convolutional neural networks is based on the pre-processed training images and the labels associated with the training images.
10. An apparatus for classifying images using one or more image classification models, comprising:
processing circuitry configured to
obtain training images associated with labels, wherein at least one training image of the training images is associated with two or more labels of the labels, each label corresponding to an image classification class, the labels having a hierarchical structure,
train at least two convolutional neural networks using the training images and the hierarchically-structured labels associated with the training images, a separate convolutional neural network being trained for each level of the hierarchical structure, and
classify, by multiplying a probability score output from a first trained convolutional neural network for a label by a probability score output from a second trained convolutional neural network for the label, an input image into two or more classes based on the trained at least two convolutional neural networks.
11. The apparatus according to claim 10 , wherein a classification layer of each respective convolutional neural network is based on soft-sigmoid activation, the soft-sigmoid activation being a combination of a softmax function and a sigmoid function.
12. The apparatus according to claim 10 , wherein the training images and the input images include graphically-designed images.
13. The apparatus according to claim 10 , wherein the labels are non-mutually exclusive labels.
14. The apparatus according to claim 10 , wherein the labels are codes used by a trademark registration organization.
15. The apparatus according to claim 10 , wherein the processing circuitry is further configured to
pre-process the training images, the training the at least two convolutional neural networks being based on the pre-processed training images and the labels associated with the training images.
16. A non-transitory computer-readable storage medium storing computer-readable instructions that, when executed by a computer, cause the computer to perform a method for classifying images using one or more image classification models, comprising:
obtaining training images associated with labels, wherein at least one training image of the training images is associated with two or more labels of the labels, each label corresponding to an image classification class, the labels having a hierarchical structure;
training at least two convolutional neural networks using the training images and the hierarchically-structured labels associated with the training images, a separate convolutional neural network being trained for each level of the hierarchical structure; and
classifying, by multiplying a probability score output from a first trained convolutional neural network for a label by a probability score output from a second trained convolutional neural network for the label, an input image into two or more classes based on the trained at least two convolutional neural networks.
17. The non-transitory computer-readable storage medium according to claim 16 , wherein a classification layer of each respective convolutional neural network is based on soft-sigmoid activation, the soft-sigmoid activation being a combination of a softmax function and a sigmoid function.
18. The non-transitory computer-readable storage medium according to claim 16 , wherein the training images and the input images include graphically-designed images.
19. The non-transitory computer-readable storage medium according to claim 16 , wherein the labels are non-mutually exclusive labels.
20. The non-transitory computer-readable storage medium according to claim 16 , wherein the labels are codes used by a trademark registration organization.