EUVIMEDEUVIMED entdecken
Crossref

DepthShield: Optical Depth Encoding for Image Tampering Detection

Jhon Lopez, Edwin Vargas, Junzhe Wang, Salman Khan, Henry Arguello, Ashok Veeraraghavan

IEEE Transactions on Pattern Analysis and Machine Intelligence · 2026 · 1-12

DOI: 10.1109/tpami.2026.3750903 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026. Verfasst von Jhon Lopez und 5 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Detecting image forgeries is increasingly difficult as diffusion-based and manual edits preserve perceptually plausible image statistics. Most existing tampering detection methods rely solely on post-capture data, assuming a trusted acquisition process, and thus degrade when attackers replicate the image formation distribution. We introduce DepthShield, a depth assisted integrity framework that enforces authenticity at the moment of capture by coupling optical encoding with learned verification. The system operates in two coordinated stages that separate secure capture from high-quality release to end users. In the Image Capture Stage (IC-Stage), a designed phase mask modulates the image formation process to encode depth dependent signatures. This establishes a deterministic mapping between scene geometry and the encoded image, prioritizing security over initial visual quality (∼ 20 dB). Any unauthorized manipulation disrupts this fundamental RGB-Depth relationship. In the Integrity Restoration Stage (IR-Stage), a high-quality image (∼30 dB) is reconstructed from the protected capture and its depth cues while anchoring structural depth relationships necessary for reliable authentication. Rather than relying on statistical RGB artifacts only, DepthShield verifies whether an image is authentic by verifying the geometric consistency between the recovered RGB intensities and its corresponding depth map, as defined by the inverse mapping of the optical system. Experiments demonstrate that DepthShield surpasses software-only baselines across diverse tampering types, including diffusion-based edits, splicing, and local edits, and remains robust under common degradations such as compression. In addition, we implement a hardware prototype, validating its effectiveness on real-world captured scenes.

Zitieren

APA

Lopez, J., Vargas, E., Wang, J., Khan, S., Arguello, H., & Veeraraghavan, A. (2026). DepthShield: Optical Depth Encoding for Image Tampering Detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1-12. https://doi.org/10.1109/tpami.2026.3750903

Vancouver

Lopez J, Vargas E, Wang J, Khan S, Arguello H, Veeraraghavan A. DepthShield: Optical Depth Encoding for Image Tampering Detection. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2026:1-12. doi:10.1109/tpami.2026.3750903

BibTeX
@article{Lopez2026,
  author = {Lopez, Jhon and Vargas, Edwin and Wang, Junzhe and Khan, Salman and Arguello, Henry and Veeraraghavan, Ashok},
  title = {DepthShield: Optical Depth Encoding for Image Tampering Detection},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {2026},
  pages = {1--12},
  doi = {10.1109/tpami.2026.3750903}
}
Kostenlos starten →