This website requires JavaScript.

TriPINet: Tripartite Progressive Integration Network for Image Manipulation Localization

Wei-Yun LiangJing XuXiao Jin
Dec 2022
摘要
Image manipulation localization aims at distinguishing forged regions fromthe whole test image. Although many outstanding prior arts have been proposedfor this task, there are still two issues that need to be further studied: 1)how to fuse diverse types of features with forgery clues; 2) how toprogressively integrate multistage features for better localizationperformance. In this paper, we propose a tripartite progressive integrationnetwork (TriPINet) for end-to-end image manipulation localization. First, weextract both visual perception information, e.g., RGB input images, and visualimperceptible features, e.g., frequency and noise traces for forensic featurelearning. Second, we develop a guided cross-modality dual-attention (gCMDA)module to fuse different types of forged clues. Third, we design a set ofprogressive integration squeeze-and-excitation (PI-SE) modules to improvelocalization performance by appropriately incorporating multiscale features inthe decoder. Extensive experiments are conducted to compare our method withstate-of-the-art image forensics approaches. The proposed TriPINet obtainscompetitive results on several benchmark datasets.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答