空天预警研究学报2025，Vol.39Issue(6)：425-430,6.DOI:10.3969/j.issn.2097-180X.2025.06.007
一种基于渐进式特征融合的图像篡改定位方法
A progressive feature fusion-based method for image manipulation localization
朱冠宇1
作者信息
- 1. 国防大学军事文化学院,北京 100081
- 折叠
摘要
Abstract
Aimed at the problems of semantic conflict and detail loss during the fusion of high-level semantic and low-level detail features in image manipulation localization,this paper proposes a progressive feature fusion method.In this method,based on a dual-stream(RGB and noise)backbone network and inspired by the idea of feature pyramid network(FPN)is designed a top-down information propagation pathway,which progressively and smoothly integrates high-level semantic information into low-level detail features to alleviate the semantic gap.Ablation experiments demonstrate that,compared with the traditional baseline model which uses only high-level features and the simple cross-layer fusion method,the proposed method can more effectively resolve the problem of multi-level feature fusion conflicts,enhance the feature representation of the manipulated areas,and thereby improve the accuracy of manipulation localization.关键词
图像篡改定位/图像取证/特征融合/图像分割/深度学习Key words
image manipulation localization/image forensics/feature fusion/image segmentation/deep learning分类
信息技术与安全科学引用本文复制引用
朱冠宇..一种基于渐进式特征融合的图像篡改定位方法[J].空天预警研究学报,2025,39(6):425-430,6.
