Multi-domain image conversion method and system based on generative adversarial network

Abstract

The invention discloses a multi-domain image conversion method and system based on a generative adversarial network. The multi-domain image conversion method comprises the steps of inputting an original image x and an original image y of a specified X mode and a specified Y mode; performing encoding and decompression on the original image x and the original image y in the reconstruction training part to obtain original image characteristics, reconstructed images and reconstructed characteristics respectively, and performing modal discriminant adversarial learning on the characteristics and the images; enabling the loop training part to generate a reconstructed graph, reconstructed graph features and a loop reconstructed graph based on an encoder of an original graph feature exchange modeof a preamble, performing modal discrimination confrontation learning of the features and the graph again, and finally outputting the loop reconstructed graph. A semi-supervised learning method is adopted, existing label data can be used, label-free data can also be used, multi-directional multi-domain image conversion can be achieved without being limited to one-way domain conversion or two-way two-domain conversion, the number of domains is not limited, and the problems of image style migration, medical image multi-mode conversion and the like can be solved.

Description

technical field [0001] The invention relates to image generation technology in the field of deep learning, in particular to a multi-domain image conversion method and system based on a generative confrontation network. Background technique [0002] In recent years, Convolution Neural Network (CNN) has shown excellent performance in computer vision, natural language processing, medical image processing and other fields, and deep learning represented by it has quickly become the core of current artificial intelligence technology research. . In 2014, the emergence of Generative Adversarial Network (GAN for short) also brought new ideas for deep learning. In 2016, GAN was combined with CNN. Since then, GAN has been widely used in many computer vision tasks. [0003] GAN is a training framework consisting of two parts, a Generator and a Discriminator, and the Generator and the Discriminator are in an Adversarial relationship. The principle of GAN is as the name suggests. The ge...

AI Technical Summary

Problems solved by technology

However, in the current popular medical image public datasets, we found that few samples are common, and paired multimodal image data is even more scarce.

Images