Image editing method, device and equipment based on topological guidance and visual feature embedding

By combining a topology graph generator and a diffusion model, high-quality, target-specific images can be generated under small sample conditions, solving the quality and controllability problems of image editing in existing technologies, and making it suitable for a variety of application scenarios.

CN122336065APending Publication Date: 2026-07-03INST OF AUTOMATION CHINESE ACAD OF SCI
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
INST OF AUTOMATION CHINESE ACAD OF SCI
Filing Date
2026-06-04
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing image editing techniques struggle to generate high-quality, semantically rich, and structurally complex images of specific targets, and they also fail to meet the editing needs of small samples and high customization. Traditional methods are prone to introducing irrelevant noise, and generative adversarial networks face bottlenecks in terms of training stability and controllability of generated content.

Method used

This paper proposes an image editing method that employs topology-guided editing and visual feature embedding. By combining a topology graph generator and a diffusion model with topology-guided editing and image reference editing strategies, high-quality edited images are generated. The topology-guided editing strategy fine-tunes the base model and generates a topology conditional graph, while the image reference editing strategy expands the input channels and generates semantic features from the reference image, enabling precise editing of the target object.

Benefits of technology

It generates high-quality, target-specific images under small sample conditions, meets users' editing needs, improves the quality and controllability of image editing, and is suitable for specific image synthesis and artistic element synthesis creation in closed or open environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122336065A_ABST
    Figure CN122336065A_ABST
Patent Text Reader

Abstract

This invention provides a topology-guided and visual feature embedding image editing method, apparatus, and device, relating to the field of image processing technology, to address the defect of poor quality of edited images generated during image editing. The method includes: acquiring the image to be edited and editing intent information; selecting a target editing strategy from a set of preset editing strategies based on the editing intent information, the multiple editing strategies including at least a topology-guided editing strategy and an image reference editing strategy; invoking a diffusion model corresponding to the target editing strategy, and generating an edited target image based on the image to be edited and the editing intent information; fine-tuning a base model based on multiple samples containing the target object to obtain a diffusion model corresponding to the topology-guided editing strategy; and expanding the input channels of the base model to obtain a diffusion model corresponding to the image reference editing strategy, the expanded channels being used to input a reference image including the target object.
Need to check novelty before this filing date? Find Prior Art