Learning device, image processing device, parameter generation device, learning method, and image processing method

An image processing device comprising:

• first processing circuitry configured to
• generate a third image quality parameter by using a second learned model generated by a learning device,
• process a processing target image by using the generated third image quality parameter,
• acquire a first processed image obtained by processing the processing target image with the third image quality parameter,
• process the processing target image by using a fourth image quality parameter having a predetermined value, and acquire a second processed image obtained by processing the processing target image by using the fourth image quality parameter,
• use the first learned model to acquire a third evaluation result for the first processed image,
• use the first learned model to acquire a fourth evaluation result for the second processed image, and
• select the first processed image when the third evaluation result is equal to the fourth evaluation result or when the third evaluation result is better than the fourth evaluation result, and select the second processed image when the fourth evaluation result is better than the third evaluation result,
• the learning device including second processing circuitry configured to:
• perform first machine learning using first training data including a first evaluation result for an evaluation target image to generate a first learned model outputting a second evaluation result for an input image;
• use the first learned model to acquire a plurality of the second evaluation results for a plurality of the input images;
• select a second image quality parameter from a plurality of first image quality parameters having different values, based on the plurality of the second evaluation results, and generate second training data including the selected second image quality parameter;
• and
• perform second machine learning using the second training data to generate the second learned model outputting the third image quality parameter used for processing the processing target image.

A parameter generation device comprising:

• first processing circuitry configured to
• acquire, from a learning device, a second learned model generated by the learning device;
• and
• use the acquired second learned model to generate a third image quality parameter,
• acquire a first processed image obtained by processing the processing target image with the third image quality parameter,
• process the processing target image by using a fourth image quality parameter having a predetermined value, and acquire a second processed image obtained by processing the processing target image by using the fourth image quality parameter,
• use the first learned model to acquire a third evaluation result for the first processed image,
• use the first learned model to acquire a fourth evaluation result for the second processed image, and
• select the first processed image when the third evaluation result is equal to the fourth evaluation result or when the third evaluation result is better than the fourth evaluation result, and select the second processed image when the fourth evaluation result is better than the third evaluation result,
• the learning device including second processing circuitry configured to:
• perform first machine learning using first training data including a first evaluation result for an evaluation target image to generate a first learned model outputting a second evaluation result for an input image;
• use the first learned model to acquire a plurality of the second evaluation results for a plurality of the input images;
• select a second image quality parameter from a plurality of first image quality parameters having different values, based on the plurality of the second evaluation results, and generate second training data including the selected second image quality parameter;
• and
• perform second machine learning using the second training data to generate the second learned model outputting the third image quality parameter used for processing the processing target image.

A learning method comprising:

• performing first machine learning using first training data including a first evaluation result for an evaluation target image to generate a first learned model outputting a second evaluation result for an input image;
• using the first learned model to acquire a plurality of the second evaluation results for a plurality of the input images;
• selecting a second image quality parameter from a plurality of first image quality parameters having different values, based on the plurality of the second evaluation results;
• generating second training data including the selected second image quality parameter;
• performing second machine learning using the second training data to generate a second learned model outputting a third image quality parameter used for processing a processing target image, wherein a first processed image is obtained by an image processing device by processing the processing target image with the third image quality parameter and a second processed image is obtained by the image processing device by processing the processing target image by using a fourth image quality parameter having a predetermined value;
• generating a third evaluation result for the first processed image;
• and
• generating a fourth evaluation result for the second processed image,
• wherein the first processed image is selected by the image processing device when the third evaluation result is equal to the fourth evaluation result or when the third evaluation result is better than the fourth evaluation result, and the second processed image is selected by the image processing device when the fourth evaluation result is better than the third evaluation result.

An image processing method comprising:

• generating a third image quality parameter by using a second learned model generated by a learning device;
• and
• processing a processing target image by using the generated third image quality parameter,
• acquiring a first processed image obtained by processing the processing target image with the third image quality parameter,
• processing the processing target image by using a fourth image quality parameter having a predetermined value, and acquire a second processed image obtained by processing the processing target image by using the fourth image quality parameter,
• using the first learned model to acquire a third evaluation result for the first processed image,
• using the first learned model to acquire a fourth evaluation result for the second processed image, and
• selecting the first processed image when the third evaluation result is equal to the fourth evaluation result or when the third evaluation result is better than the fourth evaluation result, and select the second processed image when the fourth evaluation result is better than the third evaluation result,
• performing a first machine learning using first training data including a first evaluation result for an evaluation target image to generate a first learned model outputting a second evaluation result for an input image;
• using the first learned model to acquire a plurality of the second evaluation results for a plurality of the input images;
• selecting a second image quality parameter from a plurality of first image quality parameters having different values, based on the plurality of the second evaluation results, and generating second training data including the selected second image quality parameter;
• and
• performing second machine learning using the second training data to generate the second learned model outputting the third image quality parameter used for processing the processing target image.