Method for processing image for improving the quality of the image and apparatus for performing the same

A method of training an artificial neural network, which is run on a computing device and used to create a class map for improving image quality, the method comprising:

• creating a ground truth frequency map for each of the frequency channels by applying a conversion function to a training ground truth image;
• converting the ground truth frequency map into a ground truth class map;
• inputting a training image corresponding to the training ground truth image to the artificial neural network as an input variable to create a training estimated class probability distribution map for each of the frequency channels;
• determining a class for each of the frequency channels based on a probability distribution included in the training estimated class probability distribution map and creating a training estimated class map that includes the determined class;
• and
• training the artificial neural network in order to minimize loss between the training estimated class map and the ground truth class map.

A computing device that processes an image to improve image quality, comprising:

• an encoder configured to input the image to a first artificial neural network as an input variable to create a feature map;
• a frequency processing module configured to receive the image and create an estimated frequency map based on the image;
• a concatenating module configured to concatenate the estimated frequency map with the feature map to create a concatenated feature map;
• and
• a decoder configured to input the concatenated feature map to a third artificial neural network as an input variable to create a restored image,
• wherein the frequency processing module includes:
• a classifier configured to input the image to a second artificial neural network as an input variable to create an estimated class probability distribution map for each of the frequency channels;
• a class determination module configured to determine a class for each of the frequency channels based on a probability distribution included in the estimated class probability distribution map to create an estimated class map that includes the determined class;
• and
• a class-to-coefficient conversion module configured to convert the estimated class map into the estimated frequency map.