Method of color decomposition and method of demosaicing images based on deep learning using the same

A method of demosaicing images, comprising:

• generating inter-color images indicating similarity between color sensitivities based on color images;
• determining conversion coefficients of the color images and the inter-color images with respect to a white image;
• generating a pseudo-white image corresponding to the color images and the inter-color images using the conversion coefficients;
• generating training mosaic images based on the color images and the pseudo-white image;
• training an artificial neural network based on the training mosaic images;
• and
• generating demosaiced color images corresponding to input mosaic images based on the artificial neural network that is trained.

A method of demosaicing images, comprising:

• generating a green-blue image indicating similarity between green sensitivity and blue sensitivity and generating a green-red image indicating similarity between green sensitivity and red sensitivity, based on a red image, a green image, and a blue image;
• determining conversion coefficients of the red image, the green image, the blue image, the green-blue image, and the green-red image with respect to a white image;
• generating a pseudo-white image corresponding to the red image, the green image, the blue image, the green-blue image, and the green-red image using the conversion coefficients;
• generating training mosaic images based on the red image, the green image, the blue image, and the pseudo-white image;
• training an artificial neural network based on the training mosaic images;
• and
• generating demosaiced color images corresponding to input mosaic images based on the artificial neural network that is trained.