Image synthesis method and system combining adversarial auto-encoder and generative adversarial network

Abstract

The invention discloses an image synthesis method and system combining an adversarial auto-encoder and a generative adversarial network. The method comprises the following steps: constructing an enhanced confrontation automatic encoder comprising two groups of encoders of different categories, two groups of first discriminators of different categories and a group of decoders; constructing an improved conditional generative adversarial network comprising a generator and a second discriminator; taking the artificially segmented blood vessel tree image and the original fundus retina image as training data, and performing iterative training on the enhanced adversarial autoencoder and the improved conditional generative adversarial network to obtain an optimal blood vessel tree image generatorand an optimal fundus retina image generator; and performing fundus retina image synthesis on the to-be-processed artificially segmented blood vessel tree image based on the optimal blood vessel treeimage generator and the optimal fundus retina image generator to obtain a synthesized image. According to the invention, sample data with higher precision and more diversified styles can be generated,and limited training sample data can be effectively amplified.

Description

technical field [0001] The invention relates to the field of image processing, in particular to an image synthesis method and system combining an adversarial self-encoder and a generative adversarial network. Background technique [0002] In the field of medical applications, medical image processing algorithms with superior performance require a large amount of effective medical image data with specific annotation information as training sample data. In the actual treatment process, medical equipment will have a large amount of radiation that will damage human health and involve the patient's own privacy, so it is difficult to directly obtain medical image data, so medical image data samples of different organs of the human body are very scarce; at the same time, different The lesions caused by human organs are of different complexity, and it is very difficult and expensive to manually label effective feature information labels. Therefore, in order to promote the improveme...

AI Technical Summary

Problems solved by technology

[0005] In summary, although the traditional medical data augmentation method can increase the limited data set in quantity, the augmented image lacks diversity changes; although the medical image augmentation method based on deep learning can increase the number and diversity In general, limited training samples are amplified, but the training process of the synthetic network lacks stability, and the synthetic model needs to be further optimized

Images