Infrared and visible image fusion method based on base-detail decoupling and residual band fusion

By employing a method of substrate-detail decoupling and residual band fusion, infrared and visible light image features are explicitly decomposed and differentiated, solving the problem of coupling between structural and detail information processing in existing technologies and generating fused images with higher stability and naturalness.

CN122415345APending Publication Date: 2026-07-17CHENGDU UNIVERSITY OF TECHNOLOGY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHENGDU UNIVERSITY OF TECHNOLOGY
Filing Date
2026-04-10
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

In existing infrared and visible light image fusion methods, the coupling of structural information and detail information processing leads to problems such as brightness drift, texture artifacts, noise diffusion, and insufficient modal complementarity. Existing deep learning methods struggle to achieve a controllable balance between preserving infrared saliency and maintaining visible light texture.

Method used

By constructing a feature-decoupled encoder, image information is explicitly decomposed into basal and detail features. Basis frequency band fusion layer and detail frequency band fusion layer are designed separately for differentiated processing, realizing the division of labor and fusion and collaborative optimization of structure and detail. The basal frequency band fusion layer is used to perform adaptive weighting of low-frequency subbands and controlled weighting of high-frequency subbands, while the detail frequency band fusion layer performs residual frequency band packing transformation and attention competition fusion.

Benefits of technology

It generates structurally stable, detailed, and visually natural fused images, avoiding brightness distortion and noise diffusion, and maintaining the clarity of the outlines of prominent infrared targets and visible light backgrounds, making it suitable for a variety of application scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122415345A_ABST
    Figure CN122415345A_ABST
Patent Text Reader

Abstract

本发明公开了一种基于基底‑细节解耦与残差频带融合的红外与可见光图像融合方法,包括获取配准的红外与可见光图像并预处理;通过包含共享主干与私有分支的编码器提取并解耦出相应模态的基底特征与细节特征;基底频带融合层对各模态基底特征进行子带分解,对低频和高频子带分别执行自适应加权融合和受控加权融合后重构得到融合基底特征;细节频带融合层利用残差频带打包变换获取残差增强的细节频带表征,并通过注意力机制进行自适应竞争融合得到融合细节特征;最后解码重构输出融合图像。本发明通过显式解耦与差异化频带融合策略,有效解决了融合过程中结构漂移、细节丢失与噪声扩散的问题,实现了结构稳定、细节清晰、视觉自然的融合图像生成。
Need to check novelty before this filing date? Find Prior Art