DE-FIVE: Detecting Malicious Image Prompts via Fourier Features and Image Vector Embeddings

Abstract

Vision language models (VLMs) employ both visual and textual modalities to enable advanced vision-language inference. However, incorporating visual modalities expands the attack surface of VLMs, making them more susceptible to security threats such as adversarial perturbations and indirect prompt injection, wherein crafted malicious image prompts can elicit unintended model outputs. Existing defense methods against malicious image prompts remain insufficient as they typically demand extensive da

Categories

Framework mappings

Suggested from the entry's categories.

@misc{zhong2026defive,
  title = {{DE-FIVE: Detecting Malicious Image Prompts via Fourier Features and Image Vector Embeddings}},
  author = {Xingwei Zhong and Varun Sharma and Kar Wai Fok and Vrizlynn L. L. Thing},
  year = {2026},
  month = jun,
  eprint = {2606.22779},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2606.22779}
}