Privacy-Shielded Image Compression: Defending Against Exploitation from Vision-Language Pretrained Models
On the map: Multimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness
Abstract (arXiv)
The improved semantic understanding of vision-language pretrained (VLP) models has made it increasingly difficult to protect publicly posted images from being exploited by search engines and other similar tools. In this context, this paper seeks to protect users' privacy by implementing defenses at the image compression stage to prevent exploitation. Specifically, we propose a flexible coding method, termed Privacy-Shielded Image Compression (PSIC), that can produce bitstreams with multiple decoding options. By default, the bitstream is decoded to preserve satisfactory perceptual quality while preventing interpretation by VLP models. Our method also retains the original image compression functionality. With a customizable input condition, the proposed scheme can reconstruct the image that preserves its full semantic information. A Conditional Latent Trigger Generation (CLTG) module is proposed to produce bias information based on customizable conditions to guide the decoding process into different reconstructed versions, and an Uncertainty-Aware Encryption-Oriented (UAEO) optimization function is designed to leverage the soft labels inferred from the target VLP model's uncertainty on the training data. This paper further incorporates an adaptive multi-objective optimization strategy to obtain improved encrypting performance and perceptual quality simultaneously within a unified training process. The proposed scheme is plug-and-play and can be seamlessly integrated into most existing Learned Image Compression (LIC) models. Extensive experiments across multiple downstream tasks have demonstrated the effectiveness of our design.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 32 of its 46 references that are in the map. First posted on arXiv 18 Jun 2025.
Builds on
32 of 46 references.
Closest ancestors · 1
Same direction, newest first
Other references · 31
- 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
- 2023BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Modelscited by 6,137
- 2022BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generationcited by 4,322
- 2021Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervisioncited by 3,864
- 2023InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuningcited by 2,876
- 2015VQA: Visual Question Answeringcited by 2,506
- 2015Microsoft COCO Captions: Data Collection and Evaluation Servercited by 1,462
- 2019VisualBERT: A Simple and Performant Baseline for Vision and Languagecited by 1,296
- 2017BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chaincited by 1,106
- 2023Kosmos-2: Grounding Multimodal Large Language Models to the Worldcited by 1,089
- 2018Variational image compression with a scale hyperpriorcited by 935
- 2016End-to-end Optimized Image Compressioncited by 696
19 more
- 2021CLIP4Clip: An Empirical Study of CLIP for End to End Video Clip Retrievalcited by 651
- 2020Invisible Backdoor Attack with Sample-Specific Triggerscited by 353
- 2022X-CLIP: End-to-End Multi-grained Contrastive Learning for Video-Text Retrievalcited by 280
- 2020Channel-wise Autoregressive Entropy Models for Learned Image Compressioncited by 264
- 2019A new Backdoor Attack in CNNs by training set corruption without label poisoningcited by 242
- 2021CLIP2Video: Mastering Video-Text Retrieval via Image CLIPcited by 225
- 2021A Simple Baseline for Open-Vocabulary Semantic Segmentation with Pre-trained Vision-language Modelcited by 223
- 2022A Survey of Vision-Language Pre-Trained Modelscited by 177
- 2022Lossy Image Compression with Conditional Diffusion Modelscited by 170
- 2017Real-Time Adaptive Image Compressioncited by 130
- 2023BadCLIP: Dual-Embedding Guided Backdoor Attack on Multimodal Contrastive Learningcited by 115
- 2022CLIP-Driven Fine-grained Text-Image Person Re-identificationcited by 79
- 2020Causal Contextual Prediction for Learned Image Compressioncited by 73
- 2023Generative Text-Guided 3D Vision-Language Pretraining for Unified Medical Image Segmentationcited by 47
- 2022Data Poisoning Attacks Against Multimodal Encoderscited by 41
- 2023Backdoor Attacks Against Deep Image Compression via Adaptive Frequency Triggercited by 40
- 2023CDUL: CLIP-Driven Unsupervised Learning for Multi-Label Image Classificationcited by 39
- 2022SgVA-CLIP: Semantic-guided Visual Adapting of Vision-Language Models for Few-shot Image Classificationcited by 34
- 2024Robust and Transferable Backdoor Attacks Against Deep Image Compression With Selective Frequency Priorcited by 11
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2025On the Robustness of Vision-Language Models in Zero-shot Privacy Classification
- 2024The Phantom Menace: Unmasking Privacy Leakages in Vision-Language Models
- 2026The Vision Encoder as a Privacy Boundary: Visual-Token Side Channels in Encoder-Free Vision-Language Models
- 2025Beyond Blanket Masking: Examining Granularity for Privacy Protection in Images Captured by Blind and Low Vision Users
- 2025VLIC: Vision-Language Models As Perceptual Judges for Human-Aligned Image Compression
- 2026Vision Language Model Helps Private Information De-Identification in Vision Data
- 2026When Visual Privacy Protection Meets Multimodal Large Language Models
- 2024PID: Prompt-Independent Data Protection Against Latent Diffusion Models