Image processing apparatus, image processing method, and storage medium

1. An image processing apparatus that executes noise reduction processing for an image by using multiple neural networks subjected to learning of results obtained by executing the noise reduction processing on images each having a predetermined noise level, the apparatus comprising:

a retaining unit configured to retain combination data showing a combination of multiple pieces of the noise reduction processing for each of multiple noise levels, the combination data associating a parameter for the multiple noise levels with at least either a combination of neural networks to be applied in the multiple neural networks or a number of applications of the neural networks;

an obtaining unit configured to obtain information on the noise levels from a processing target image;

a determination unit configured to determine, based on the obtained information on the noise levels, a combination of pieces of the noise reduction processing for the processing target image from the combination data retained by the retaining unit; and

an execution unit configured to execute, based on data on the combination of pieces of the noise reduction processing determined by the determination unit, the noise reduction processing using at least one of the multiple neural networks for the processing target image,

wherein combinations of multiple pieces of the noise reduction processing include a pluralities of combinations differing among a pluralities of cases where image quality is prioritized and a pluralities of cases where speed is prioritized.

2. The image processing apparatus according to claim 1 , wherein

the parameter for the multiple noise levels is information on an ISO sensitivity in capturing the image, a gain value for sensitivity correction, or a variance value of noise.

3. The image processing apparatus according to claim 1 , wherein

the number of applications of the neural networks is set so that the neural networks are applied multiple times to an image having the predetermined noise level.

4. The image processing apparatus according to claim 1 , wherein the processing target image is a partial image obtained by segmenting an input image, the apparatus further comprising:

a segmentation unit configured to segment the input image into partial images; and

an integration unit configured to integrate the partial images on which the noise reduction processing has been executed.

5. The image processing apparatus according to claim 4 , wherein

the input image is a still image, and

each of the partial images is part of the still image.

6. The image processing apparatus according to claim 4 , wherein

the input image is a moving image, and

each of the partial images is one frame of the moving image.

7. The image processing apparatus according to claim 1 , wherein the processing target image is an image of a brightness component or each of color components obtained by separating an input image, the apparatus further comprising:

a separation unit configured to separate the input image into the image of the brightness component and the image of each of the color components; and

an integration unit configured to integrate the image of the brightness component and the image of each of the color components on which the noise reduction processing has been executed.

8. The image processing apparatus according to claim 1 , further comprising:

a display unit configured to display the image on which the noise reduction processing has been executed; and

an input unit configured to input user's instructions for additional processing,

wherein the execution unit additionally executes the noise reduction processing according to the user's instructions.

9. The image processing apparatus according to claim 1 , wherein the determination unit is further configured to determine whether or not a residual noise level of the image on which the noise reduction processing has been executed satisfies predetermined conditions,

wherein in a case where it is determined that the residual noise level does not satisfy the predetermined conditions, the execution unit additionally executes the noise reduction processing.

10. The image processing apparatus according to claim 1 , wherein

the combination of multiple pieces of the noise reduction processing in the case where the image quality is prioritized is a combination in which the noise reduction processing relatively less likely to cause image quality deterioration is executed more times than in the case where the speed is prioritized.

11. An image processing method of executing noise reduction processing for an image by using multiple neural networks subjected to learning of results obtained by executing the noise reduction processing on images each having a predetermined noise level, the method comprising:

obtaining information on a noise level from a processing target image;

determining, based on the obtained information on the noise level, a combination of pieces of the noise reduction processing for the processing target image from combination data showing a combination of multiple pieces of the noise reduction processing for each of multiple noise levels, the combination data associating a parameter for the multiple noise levels with at least either a combination of neural networks to be applied in the multiple neural networks or a number of applications of the neural networks; and

executing, based on data on the determined combination of pieces of the noise reduction processing, the noise reduction processing using at least one of the multiple neural networks for the processing target image,

wherein combinations of multiple pieces of the noise reduction processing include a pluralities of combinations differing among a pluralities of cases where image quality is prioritized and a pluralities of cases where speed is prioritized.

12. The image processing method according to claim 11 , wherein

the parameter for the multiple noise levels is information on an ISO sensitivity in capturing the image, a gain value for sensitivity correction, or a variance value of noise.

13. The image processing method according to claim 11 , wherein

the number of applications of the neural networks is set so that the neural networks are applied multiple times to an image having the predetermined noise level.

14. The image processing method according to claim 11 , wherein the processing target image is a partial image obtained by segmenting an input image, the method further comprising:

segmenting the input image into partial images; and

integrating the partial images on which the noise reduction processing has been executed.

15. The image processing method according to claim 14 , wherein

the input image is a still image, and

each of the partial images is part of the still image.

16. The image processing method according to claim 14 , wherein

the input image is a moving image, and

each of the partial images is one frame of the moving image.

17. The image processing method according to claim 11 , wherein the processing target image is an image of a brightness component or each of color components obtained by separating an input image, and

wherein the method further comprises:

separating the input image into the image of the brightness component and the image of each of the color components; and

integrating the image of the brightness component and the image of each of the color components on which the noise reduction processing has been executed.

18. A non-transitory computer readable storage medium storing a program for causing a computer to perform a method of executing noise reduction processing for an image having a predetermined noise level by using multiple neural networks subjected to learning of results obtained by executing the noise reduction processing, the method comprising:

obtaining information on a noise level from a processing target image;

determining, based on the obtained information on the noise level, a combination of pieces of the noise reduction processing for the processing target image from combination data showing a combination of multiple pieces of the noise reduction processing for each of multiple noise levels, the combination data associating a parameter for the multiple noise levels with at least either a combination of neural networks to be applied in the multiple neural networks or a number of applications of the neural networks; and

executing, based on data on the determined combination of pieces of the noise reduction processing, the noise reduction processing using at least one of the multiple neural networks for the processing target image,

wherein combinations of multiple pieces of the noise reduction processing include a pluralities of combinations differing among a pluralities of cases where image quality is prioritized and a pluralities of cases where speed is prioritized.