Learning method, storage medium and image processing device

1. A learning method for causing a statistical model to learn, the statistical model being generated by learning a bokeh caused in a first image captured in a first domain in accordance with a distance to a first subject included in the first image, the method comprising:

acquiring a plurality of second images by capturing a second subject from multiple viewpoints in a second domain other than the first domain; and

causing the statistical model to learn using the second images.

2. The learning method of claim 1 , further comprising:

obtaining a distance to the second subject from each of the second images; and

converting the distance obtained from each of the second images into a bokeh value indicating a bokeh caused in accordance with the distance to the second subject,

wherein the causing the statistical model to learn comprises causing the statistical model to learn the second images and the bokeh value.

3. The learning method of claim 2 , further comprising calculating,

wherein:

the distance obtained from each of the second images is a distance with an indefinite scale;

the calculating comprises matching the distance obtained from each of the second images with a bokeh value output from the statistical model by inputting each of the second images to the statistical model to calculate a first parameter for converting the distance into a distance based on an actual scale and a second parameter that is indefinite in a capture device that has captured the second images; and

the converting comprises converting the distance obtained from each of the second images into a bokeh value indicating a bokeh caused in accordance with the distance, using the first parameter and the second parameter.

4. The learning method of claim 3 , wherein the calculating comprises calculating the first parameter and the second parameter by linearly regressing a reciprocal of the distance obtained from each of the second images and a bokeh value output from the statistical model by inputting each of the second images to the statistical model.

5. The learning method of claim 3 , wherein the calculating comprises calculating the first parameter and the second parameter by nonlinear optimization based on the distance obtained from each of the second images and a bokeh value output from the statistical model by inputting each of the second images to the statistical model.

6. The learning method of claim 5 , wherein the calculating includes calculating initial values of the first parameter and the second parameter by linearly regressing a reciprocal of the distance obtained from each of the second images and a bokeh value output from the statistical model by inputting each of the second images to the statistical model, and calculating the first parameter and the second parameter using the initial values.

7. The learning method of claim 2 , further comprising:

extracting a feature point of the second subject included in each of the second images; and

outputting an area in which the number of feature points of the second subject is less than a predetermined value in each of the second images,

wherein the obtaining the distance comprises obtaining a distance to the second subject for each of the feature points of the second subject.

8. The learning method of claim 1 , further comprising acquiring a third image with a known distance from the third image to a third subject,

wherein the causing the statistical model to learn comprises causing the statistical model to learn using the second images and the third image.

9. A non-transitory computer-readable storage medium having stored thereon a computer program which is executable by a computer and causes a statistical model to learn, the statistical model being generated by learning a bokeh caused in a first image captured in a first domain in accordance with a distance to a first subject included in the first image, the computer program comprising instructions capable of causing the computer to execute functions of:

acquiring a plurality of second images by imaging a second subject from multiple viewpoints in a second domain other than the first domain; and

causing the statistical model to learn the bokeh using the second images.

10. An image processing device which causes a statistical model to learn, the statistical model being generated by learning a bokeh caused in a first image captured in a first domain in accordance with a distance to a first subject included in the first image, the device comprising a processor configured to:

acquire a plurality of second images by imaging a second subject from multiple viewpoints in a second domain other than the first domain; and

cause the statistical model to learn the bokeh using the second images.