Image processing device, image processing method, and image processing program

1. An image processing device comprising:

at least one memory operable to store program code; and

at least one processor configured to read the program code and operate as instructed by the program code, the program code comprising:

acquisition code configured to cause the at least one processor to acquire an input image;

generation code configured to cause the at least one processor to generate a plurality of comparison images by compressing a target region being at least part of the input image with each of a plurality of compression levels and expanding the compressed target region to its original size;

calculation code configured to cause the at least one processor to calculate, for each of the plurality of comparison images, a degradation level of the comparison image with respect to the input image;

estimation code configured to cause the at least one processor to estimate a blur level of the input image based on a plurality of calculated degradation levels, and

super-resolution code configured to cause the at least one processor to select a machine learning model corresponding to the blur level estimated by the estimation code from a plurality of machine learning models for super-resolution respectively corresponding to a plurality of blur levels, and input the input image into the selected machine learning model to generate a high-resolution image.

2. The image processing device according to claim 1 , wherein the degradation level is peak signal-to-noise ratio.

3. The image processing device according to claim 1 , wherein the estimation code is further configured to cause the at least one processor to estimate the blur level by comparing the plurality of degradation levels with a threshold.

4. The image processing device according to claim 1 , wherein the estimation code is further configured to cause the at least one processor to estimate the blur level based on a leading coefficient of a non-linear function representing a relationship between the compression level and the degradation level.

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

a correspondence between the leading coefficient and the blur level is set in advance, and

the calculation code is further configured to cause the at least one processor to calculate the leading coefficient of the non-linear function from the plurality of compression levels and the plurality of degradation levels, and determine the blur level corresponding to the calculated leading coefficient by referring to the correspondence.

6. An image processing method performed by an image processing device including a processor, comprising:

acquiring an input image;

generating a plurality of comparison images by compressing a target region being at least part of the input image with each of a plurality of compression levels and expanding the compressed target region to its original size;

calculating, for each of the plurality of comparison images, a degradation level of the comparison image with respect to the input image;

estimating a blur level of the input image based on a plurality of calculated degradation levels;

selecting a machine learning model corresponding to the blur level from a plurality of machine learning models for super-resolution respectively corresponding to a plurality of blur levels; and

inputting the input image into the selected machine learning model to generate a high-resolution image.

7. A non-transitory recording medium storing an image processing program causing a computer to:

acquire an input image;

generate a plurality of comparison images by compressing a target region being at least part of the input image with each of a plurality of compression levels and expanding the compressed target region to its original size;

calculate, for each of the plurality of comparison images, a degradation level of the comparison image with respect to the input image;

estimate a blur level of the input image based on a plurality of calculated degradation levels,

select a machine learning model corresponding to the blur level from a plurality of machine learning models for super-resolution respectively corresponding to a plurality of blur levels; and

input the input image into the selected machine learning model to generate a high-resolution image.