A texture-structure guided feature fusion and context refinement image inpainting method

CN122072948BActive Publication Date: 2026-06-26QILU UNIVERSITY OF TECHNOLOGY (SHANDONG ACADEMY OF SCIENCES)
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
QILU UNIVERSITY OF TECHNOLOGY (SHANDONG ACADEMY OF SCIENCES)
Filing Date
2026-04-21
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing image restoration methods struggle to simultaneously maintain global structural consistency and local texture fidelity when dealing with large, irregularly missing areas, resulting in structural distortion and texture blurring in the restoration results.

Method used

We employ a texture-structure guided feature fusion and contextual refinement method. By combining a texture generation network, a structure prediction network, and an image reconstruction network, we utilize a dual-branch adaptive fusion module and a multi-scale contextual refinement module to dynamically balance texture and structural features. Combined with SPADE's spatial adaptive modulation and gated convolution, we achieve robust image inpainting.

Benefits of technology

It maintains stable restoration performance within a 0%-60% occlusion range, avoids structural breakage and texture blurring, and ensures semantic coherence and visual authenticity of the restoration results, making it suitable for restoration scenarios such as murals and old photographs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122072948B_ABST
    Figure CN122072948B_ABST
Patent Text Reader

Abstract

The present application relates to a kind of texture-structure guided feature fusion and context refinement image restoration method, belong to computer vision and digital image processing technical field, to solve the problems of structure discontinuity, texture blur, structure and texture feature fusion effect is not good in existing image restoration technology.The method comprises: the restoration model including texture generation network TGN, structure prediction network SPN and image reconstruction network IRN is constructed;Complete gray scale map and edge map are generated by TGN to provide texture guide, SPN outputs global consistent structure priori map, generates initial rough feature map by double-branch adaptive fusion module, and then is optimized by the multi-scale context refinement module of IRN, and outputs final restoration image.The present application effectively improves the structure coherence and texture fidelity of restoration image.
Need to check novelty before this filing date? Find Prior Art