Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
February 24, 2026Open Access

SIDWA: Synthetic Image Detection Based on Discrete Wavelet Transform Stem and Deformable Sliding Window Cross-Attention

View Full Paper
Ask AI
Bookmark
Share

Authors

LLLuo LiTLTianyi LuJSJiaxin Song

Discussion

Loading...

Member takes

Overview

Novel dual-branch detection method identifies synthetic images with advanced frequency and spatial techniques, indicating strong forensic capabilities.

Key Points

  • To develop SIDWA, a framework for detecting synthetic images using combined spatial and frequency analysis.
  • Proposed a dual-branch detection framework combining spatial and frequency domains.
  • Implemented a Deformable Sliding Window Cross-Attention module for dynamic edge capturing.
  • Utilized Discrete Wavelet Transform Stem to decompose images into multi-scale sub-bands.
  • Applied Frequency-Semantic Resonance Projector to guide anomaly detection.
  • Achieved average accuracy exceeding 95% on the SIDataset benchmark.
  • Demonstrated competitive inference time of 18.2 ms on NVIDIA A100 GPU.
  • Ablation studies confirmed the importance of learnable offsets and frequency integration for improved robustness.

Cite This Study

Li et al. (2026) studied this question.

synapsesocial.com/papers/699d3fd9de8e28729cf64a1dhttps://doi.org/10.3390/electronics15040891
View Full Paper
Ask AI
Bookmark
Share