Training method of generative adversarial model, image processing method, device and equipment

Through the training method of generating adversarial models, using multi-scale discriminant networks to process image details, the problem of poor image repair effect in the prior art is solved, and high-resolution and high-quality image reconstruction and text information repair are achieved.

CN119722497APending Publication Date: 2025-03-28INDUSTRIAL AND COMMERCIAL BANK OF CHINA
0 Cites 0 Cited by

Patent Information

Application Number
CN202510023119.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-01-07
Publication Date
2025-03-28

AI Technical Summary

Technical Problem

The prior art has limitations in processing complex image information and retaining high-frequency details, and cannot adaptively process variable image details, resulting in a reduced repair effect of text images.

Method used

The training method of generating adversarial models is adopted, and the sample noise vector, initial image and real image in the training set are obtained, and the generated network and multi-scale discriminant network are trained. The multi-scale discriminant network is constructed based on convolutional layers of different scales, and processes samples to generate images and real images, obtain features of multiple scales, and distinguish them based on labels. The adversarial model is trained to improve image repair effect.

Benefits of technology

Through the fusion of multi-scale features, the generation of adversarial models can generate high-resolution and high-quality images with rich details and better visual effects, which enhances the realism of image reconstruction and business record text information repair.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a training method of a generative adversarial model, an image processing method, a device and equipment, which can be applied to the technical field of big data. The method comprises the steps that a training set is acquired, and the training set comprises a sample noise vector, a sample initial image, a sample real image and a sample label; inputting the sample noise vector, the sample initial image and the sample real image into a generative network to obtain a sample generated image; processing the sample generation image and the sample real image by using a plurality of cascaded convolutional layers in the multi-scale discrimination network to obtain sample generation image features and sample real image features of multiple scales; on the basis of the sample labels, the sample generation image features of the multiple scales and the sample real image features are judged, and a sample judgment result is obtained; and training the generative adversarial model based on the sample discrimination result to obtain a trained generative adversarial model.
Need to check novelty before this filing date? Find Prior Art