Related Experiment Video
Updated: Jan 15, 2026
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
1.0K
Multi-scale and deeply supervised network for image splicing localization
Sheng Qin1,2, Ce Liang1,2, Yuling Luo1,2
1Guangxi Key Lab of Brain-Inspired Computing and Intelligent Chips, School of Electronic and Information Engineering, Guangxi Normal University, Guilin, China.
Frontiers in Artificial Intelligence
|October 13, 2025
Summary
Detecting tampered images is crucial for national security. A new multi-scale network accurately locates image splicing tampering using deep supervision and global feature extraction.
Area of Science:
- Computer Vision
- Digital Forensics
- Image Processing
Background:
- Malicious image tampering poses risks to media integrity and national security.
- Existing methods for image splicing tampering localization lack sufficient global information interaction.
- Accurate detection and localization of tampered regions are necessary.
Purpose of the Study:
- To propose an effective method for localizing image splicing tampering.
- To address the limitations of existing methods by incorporating global information.
- To develop a network capable of pixel-wise prediction of tampered regions.
Main Methods:
- A multi-scale, deeply supervised image splicing tampering localization network based on an encoder-decoder architecture.
- Utilizing different levels of feature maps for deep supervision of splicing locations.
- Incorporating a multi-scale feature extraction module for expanded global view.
Main Results:
- The proposed network achieved F1 scores of 0.891 on the CASIA dataset and 0.864 on the COLUMB dataset.
- The model demonstrated accurate localization of tampered regions.
- The multi-scale feature extraction module improved the differentiation between tampered and non-tampered areas.
Conclusions:
- The proposed multi-scale, deeply supervised network is effective for image splicing tampering localization.
- The integration of global information and multi-scale features enhances detection accuracy.
- The method provides a robust solution for identifying maliciously tampered images.
