Chunqiang Yu

Deep neural networks for image classification require protection against unauthorized use and redistribution. Existing watermarking methods suffer from a critical vulnerability: watermarks are always active and detectable, allowing adversaries to identify and remove them before deployment. We propose DormMark, a novel framework for image classification models that introduces delayed-activation watermarks which remain dormant and hidden under deployment-time black-box query auditing during initial deployment, but automatically activate upon fine-tuning. Our approach employs a three-stage training paradigm: (1) embedding watermarks using triggered samples, (2) masking to suppress watermark functionality while preserving its latent presence, and (3) activation through standard fine-tuning without owner intervention. This mechanism exploits neural networks’ forgetting-remembering behaviors during continued training, creating a fragile equilibrium that behaves similarly to a clean model under deployment-time black-box auditing but reliably manifests ownership indicators after modification. We consider a private-key black-box verification setting in which the owner keeps the concrete trigger instances secret. Experiments across multiple architectures (VGG19, ResNet-18/56, DenseNet-121, WideResNet-34) and datasets (CIFAR-10, CIFAR-100, GTSRB) demonstrate 100% watermark success rates, high imperceptibility (PSNR > 38 dB, SSIM = 0.99), negligible accuracy loss (< 0.04%), and robustness against 80% parameter pruning. DormMark represents a paradigm shift from static to conditionally-activated ownership verification, providing a more robust framework for intellectual property protection.

To facilitate cross-platform access and convenient sharing of personal photos, an increasing number of consumers are uploading such content to cloud storage via various electronic devices. However, these practices notably amplify the risk of privacy breaches. Reversible data hiding in encrypted images (RDH-EI) emerges as a promising solution that secures both the original image content and the concealed secret data. This paper proposes a high-capacity RDH-EI method based on pixel prediction and entropy encoding. A key limitation of regression-based predictors is their sensitivity to outliers, which impairs parameter estimation and reduces predictive performance. Nevertheless, most existing predictors lack targeted mechanisms to mitigate this issue. To address this gap, we propose a robust regression-based predictor that effectively suppresses the adverse impact of outliers. We introduce two optimization techniques, prediction error grouping and intra-group transformation, to improve the efficiency of entropy encoding. First, a pre-trained neural network model is exploited to extract multi-channel deep features from the original image. Next, we propose a novel entropy-based channel selection mechanism to identify the most informative channels, which are then fused into a comprehensive feature. Based on this feature, prediction errors are partitioned into blocks, which are then clustered into multiple groups. Intra-group transformations, including non-negative and radix-remainder transformations, are then applied to these groups to concentrate their data distribution. Experimental results demonstrate that the proposed method achieves extremely high embedding rates across diverse test images and benchmark datasets, such as BOSSBase, BOWS2, the BrainTumor medical image dataset, and the USC-SIPI texture image dataset, outperforming a range of state-of-the-art methods.

Secure and high-capacity secret information transmission is an important task of the image hiding research. The existing image hiding methods face some critical issues: cover-based methods offer high capacity but introduce image distortion and security risks, whereas secure coverless methods have low capacity. To address these issues, this paper proposes a novel generative-based coverless multi-image hiding method called GCL-MIH, which can achieve high capacity and high security. The GCL-MIH first utilizes a feature reverse module to compress multiple secret images into multiple feature vectors and then normalizes them to generate a vector that conforms to a standard normal distribution, and finally inputs this vector into an invertible generative network (Flow-GAN) to generate a face image, enabling coverless multiple-image hiding without a predefined cover image. Experimental results demonstrate that the GCL-MIH successfully hides up to four images within a single generated face image, achieving a maximum embedding rate of 32 bpp. This capacity far exceeds those of the existing coverless methods. On the COCO test set, the generated stego images of the GCL-MIH are highly realistic (FID score: 11.98), and the recovered secret images exhibit satisfactory fidelity (the average PSNR and SSIM of four recovered secret images are 33.18 dB and 0.9412).

In a (k, n)-threshold secret image sharing (SIS) scheme, a secret image is encoded into n shadow images and distributed to the corresponding participants, enabling lossless reconstruction with any k correct shadow images. This inherent fault tolerance allows up to n–k shadow images to be lost or corrupted. However, in real-world scenarios, all shadow images are susceptible to malicious tampering, cropping, or noise during transmission and storage, making it difficult to guarantee the availability of k intact shadows. Robust secret image sharing (RSIS) schemes have been proposed to address this issue, yet existing methods often suffer significant degradation in reconstruction quality as the attack strength increases, revealing limitations in their robustness. To address these issues, we propose an RSIS scheme against malicious shadow images by Reusing Polynomial Coefficients with hash function (RSIS-RPC), which provides both malicious shadow detection and error correction capabilities. The correction capability improves with the degree of coefficient reuse, where greater reuse provides stronger resilience to pixel corruption. However, as more coefficients are reused, the size of the generated shadow images increases correspondingly, resulting in higher storage requirements. This trade-off between robustness and efficiency makes the proposed scheme adaptable to diverse application scenarios requiring secure and resilient image sharing. Experimental results and analyses demonstrate that the proposed scheme achieves superior robustness compared to existing schemes.