Processing method and processing device using same

1. A processing device for integrating a plurality of patterns included in an image for each type of pattern, comprising:

an input unit that receives an input of an image;

a first processing unit that detects each of the plurality of patterns included in the image as one of families that organize types of pattern, by subjecting the image input to the input unit to a process in a neural network trained to learn the families as training data, a number of families being smaller than a number of types of pattern;

an extraction unit that extracts a plurality of parts that include families from the image, based on a position of the families detected by the first processing unit;

a second processing unit that acquires intermediate data for an intermediate layer unique to each family, by subjecting the plurality of parts extracted by the extraction unit to a process in a neural network trained to learn the families as training data, the families organizing the types of pattern, and the number of families being smaller than the number of types of pattern;

a clustering unit that subjects the intermediate data acquired by the second processing unit to clustering in accordance with the number of types of pattern; and

a calculation unit that calculates a number of patterns included in each cluster that results from clustering in the clustering unit.

2. The processing device according to claim 1 , wherein

the second processing unit acquires intermediate data for an intermediate layer that precedes an output by one or two steps.

3. The processing device according to claim 2 , wherein

the second processing unit uses the neural network used in the first processing unit.

4. The processing device according to claim 3 , wherein

the neural network used in the first processing unit and the second processing unit includes a convolutional layer and a pooling layer and is a convolutional neural network in which a fully connected layer is excluded, and a filter in the convolutional layer in the convolutional neural network is trained to learn a processing result having a 1×1 spatial dimension.

5. The processing device according to claim 1 , wherein

the second processing unit uses the neural network used in the first processing unit.

6. The processing device according to claim 5 , wherein

the neural network used in the first processing unit and the second processing unit includes a convolutional layer and a pooling layer and is a convolutional neural network in which a fully connected layer is excluded, and a filter in the convolutional layer in the convolutional neural network is trained to learn a processing result having a 1×1 spatial dimension.

7. A processing method for integrating a plurality of patterns included in an image for each type of pattern, comprising:

receiving an input of an image;

detecting each of the plurality of patterns included in the image as one of families that organize types of pattern, by subjecting the image input to a process in a neural network trained to learn the families as training data, a number of families being smaller than a number of types of pattern;

extracting a plurality of parts that include families from the image, based on a position of the families detected;

acquiring intermediate data for an intermediate layer unique to each family, by subjecting the plurality of parts extracted to a process in a neural network trained to learn the families as training data, the families organizing the types of pattern, and the number of families being smaller than the number of types of pattern;

subjecting the intermediate data acquired to clustering in accordance with the number of types of pattern; and

calculating a number of patterns included in each cluster that results from clustering.