GAN-based Image Compression Using Mutual Information Maximizing Regularization

Recently, image compression systems based on convolutional neural networks that use flexible nonlinear analysis and synthesis transformations have been developed to improve the restoration accuracy of decoded images. A method using a framework called a generative adversarial network [1] has been reported as one of the methods aiming to improve the subjective image quality [2][3]. It optimizes the distribution of restored images to be close to that of natural images; thus it suppresses visual artifacts such as blurring, ringing, and blocking. However, since methods of this type are optimized to focus on whether the restored image is subjectively natural or not, components that are not correlated with the original image are mixed in the coding features obtained from the encoder. Thus, even though the appearance looks natural, it may be subjectively seen as a different object from the original image or the impression may be changed. In this paper, we describe a method we have developed to maximize mutual information between the coding features and the restored images. This method, which we call "regularization", makes it possible to develop image compression systems that suppress appearance differences with subjective naturalness.