Design method for correcting single-graph generative adversarial network based on cascade attention mechanism

Abstract

The invention discloses a design method of a corrected single image generative adversarial network based on a cascade attention mechanism, and belongs to the technical field of pattern recognition. On the basis of the generative adversarial network, attention information of the cascade attention mechanism is embedded and is used for an image processing task; the importance degree of channel and pixel information in the image features can be well utilized, and the learning performance of the generative adversarial network model for the image features is improved. A cascade channel attention mechanism and a cascade space attention mechanism are designed, the influence of special points in a single image on the attention mechanism can be effectively eliminated, input image features and the attention mechanism are cascaded, the stability of a generative adversarial network model and the quality of a generated image are further improved, and the method is suitable for being applied to a large-scale image. The cascaded attention mechanism constitutes a network module and is integrated in the corrected single graph generative adversarial network, so that the corrected single graph generative adversarial network can better learn complex global features of the image.

Description

technical field [0001] The invention relates to the technical field of pattern recognition, in particular to a design method of a corrected single-image generation confrontation network based on a cascade attention mechanism. Background technique [0002] Under the influence of convolutional neural networks (CNNs), generative adversarial networks (GANs) have made significant progress in image synthesis and are applicable to various image processing tasks, such as image super-resolution, image denoising, text-image synthesis, Image-to-image translation, etc. The generation confrontation network consists of two parts: the generator and the discriminator. The generator learns and simulates the distribution of real data through training, so that the input noise data distribution gradually approaches the real expected data distribution; while the discriminator is used to distinguish Whether the distribution of the data generated by the generator conforms to the distribution char...

AI Technical Summary

Problems solved by technology

However, the existing generative confrontation network has poor stabilit

Images