Related Experiment Video
Updated: May 3, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
UGEE-Net: Uncertainty-guided and edge-enhanced network for image splicing localization
Qixian Hao1, Ruyong Ren1, Shaozhang Niu2
1Beijing Key Lab of Intelligent Telecommunication Software and Multimedia, School of Computer, Beijing University of Posts and Telecommunications, Beijing 100876, China.
Summary
We developed a novel network for image splicing localization (ISL) that improves accuracy for detecting tampered images. This method enhances detection of subtle image manipulations and introduces a new dataset for training forensic models.
Area of Science:
- Computer Vision
- Digital Image Forensics
Background:
- Image splicing is a common method for image tampering, compromising authenticity.
- Existing image splicing localization (ISL) methods face limitations in accuracy, especially with imperceptible tampering and multiple manipulated regions.
Purpose of the Study:
- To introduce an advanced network for image splicing localization (ISL) that overcomes current limitations.
- To enhance the accuracy and robustness of detecting tampered images, including those with subtle manipulations.
Main Methods:
- Developed an Uncertainty-Guided and Edge-Enhanced Network (UGEE-Net) incorporating Bayesian learning for uncertainty mapping and frequency domain-based edge enhancement.
- Implemented a cross-level fusion and propagation mechanism for improved feature integration and detail enhancement.
- Introduced the HTSI12K dataset with 12,000 diverse spliced images for model training.
Main Results:
- UGEE-Net demonstrated superior detection accuracy, robustness, and generalization capabilities compared to existing methods.
- The uncertainty guidance and edge enhancement strategies synergistically boosted localization performance.
- The HTSI12K dataset provides a valuable resource for training and evaluating image forensic models.
Conclusions:
- UGEE-Net offers a significant advancement in image splicing localization, effectively addressing challenges of imperceptible tampering.
- The proposed methods and dataset contribute to the field of digital image forensics, enabling more reliable authenticity verification.
